chore: merge dev into v2 (#35591)

Co-authored-by: Frank <frank@anoma.ly>
Co-authored-by: Aarav Sareen <96787824+arvsrn@users.noreply.github.com>
Co-authored-by: Brendan Allan <git@brendonovich.dev>
Co-authored-by: opencode-agent[bot] <opencode-agent[bot]@users.noreply.github.com>
Co-authored-by: Jack <jack@anoma.ly>
Co-authored-by: Brendan Allan <14191578+Brendonovich@users.noreply.github.com>
Co-authored-by: Shoubhit Dash <shoubhit2005@gmail.com>
Co-authored-by: opencode-agent[bot] <219766164+opencode-agent[bot]@users.noreply.github.com>
Co-authored-by: James Long <longster@gmail.com>
Co-authored-by: Dustin Deus <deusdustin@gmail.com>
Co-authored-by: starptech <starptech@starptechs-MBP.fritz.box>
Co-authored-by: Luke Parker <10430890+Hona@users.noreply.github.com>
Co-authored-by: 𝓛𝓲𝓽𝓽𝓵𝓮 𝓕𝓻𝓪𝓷𝓴 <little-frank@opencord.local>
Co-authored-by: Dax <mail@thdxr.com>
Co-authored-by: usrnk1 <7547651+usrnk1@users.noreply.github.com>
Co-authored-by: Jay <53023+jayair@users.noreply.github.com>
Co-authored-by: runvip <164729189+runvip@users.noreply.github.com>
Co-authored-by: opencode <opencode@sst.dev>
Co-authored-by: Julian Coy <julian@ex-machina.co>
Co-authored-by: Vladimir Glafirov <vglafirov@gitlab.com>
Co-authored-by: Adam <2363879+adamdotdevin@users.noreply.github.com>
Co-authored-by: Kit Langton <kit.langton@gmail.com>
Co-authored-by: Simon Klee <hello@simonklee.dk>
Co-authored-by: Jay <air@live.ca>
Co-authored-by: David Hill <1879069+iamdavidhill@users.noreply.github.com>

packages/opencode/package.json

@@ -1,6 +1,6 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
-  "version": "1.17.13",
+  "version": "1.17.14",
   "name": "opencode",
   "type": "module",
   "license": "MIT",
@@ -59,7 +59,7 @@
     "@ai-sdk/amazon-bedrock": "4.0.112",
     "@ai-sdk/anthropic": "3.0.82",
     "@ai-sdk/azure": "3.0.49",
-    "@ai-sdk/cerebras": "2.0.41",
+    "@ai-sdk/cerebras": "2.0.60",
     "@ai-sdk/cohere": "3.0.27",
     "@ai-sdk/deepinfra": "2.0.41",
     "@ai-sdk/gateway": "3.0.104",
@@ -86,6 +86,7 @@
     "@openauthjs/openauth": "catalog:",
     "@opencode-ai/client": "workspace:*",
     "@opencode-ai/cli": "workspace:*",
+    "@opencode-ai/codemode": "workspace:*",
     "@opencode-ai/llm": "workspace:*",
     "@opencode-ai/plugin": "workspace:*",
     "@opencode-ai/protocol": "workspace:*",

packages/opencode/src/cli/cmd/tui.ts

@@ -132,6 +132,8 @@ export const TuiThreadCommand = cmd({
       const config = await TuiConfig.get()
 
       const network = resolveNetworkOptionsNoConfig(args)
+      const external = hasArg("--port") || hasArg("--hostname") || network.mdns === true
+      const headers = external ? ServerAuth.headers() : undefined
       const url = (await client.call("server", network)).url
 
       try {
@@ -139,6 +141,7 @@ export const TuiThreadCommand = cmd({
           url,
           sessionID: args.session,
           directory: cwd,
+          headers,
         })
       } catch (error) {
         UI.error(errorMessage(error))
@@ -154,10 +157,10 @@ export const TuiThreadCommand = cmd({
         const { Effect } = await import("effect")
         const { run } = await import("../tui/layer")
         const { createLegacyTuiPluginHost } = await import("@/plugin/tui/runtime")
-        await Effect.runPromise(
+         await Effect.runPromise(
           run({
-            client: createOpencodeClient({ baseUrl: url, directory: cwd }),
-            api: OpenCode.make({ baseUrl: url }),
+            client: createOpencodeClient({ baseUrl: url, headers, directory: cwd }),
+            api: OpenCode.make({ baseUrl: url, headers }),
             async onSnapshot() {
               const tui = writeHeapSnapshot("tui.heapsnapshot")
               const server = await client.call("snapshot", undefined)

packages/opencode/src/effect/runtime-flags.ts

@@ -45,6 +45,7 @@ export class Service extends ConfigService.Service<Service>()("@opencode/Runtime
   experimentalLspTool: enabledByExperimental("OPENCODE_EXPERIMENTAL_LSP_TOOL"),
   experimentalOxfmt: enabledByExperimental("OPENCODE_EXPERIMENTAL_OXFMT"),
   experimentalPlanMode: enabledByExperimental("OPENCODE_EXPERIMENTAL_PLAN_MODE"),
+  experimentalCodeMode: enabledByExperimental("OPENCODE_EXPERIMENTAL_CODE_MODE"),
   experimentalEventSystem: enabledByExperimental("OPENCODE_EXPERIMENTAL_EVENT_SYSTEM"),
   experimentalWorkspaces: enabledByExperimental("OPENCODE_EXPERIMENTAL_WORKSPACES"),
   experimentalIconDiscovery: enabledByExperimental("OPENCODE_EXPERIMENTAL_ICON_DISCOVERY"),

packages/opencode/src/mcp/browser.ts

@@ -0,0 +1,37 @@
+import { LayerNode } from "@opencode-ai/core/effect/layer-node"
+import { Context, Effect, Layer } from "effect"
+import open from "open"
+
+export interface Interface {
+  readonly open: (url: string) => Effect.Effect<void, Error>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/McpBrowser") {}
+
+const layer = Layer.succeed(
+  Service,
+  Service.of({
+    open: Effect.fn("McpBrowser.open")(function* (url: string) {
+      const subprocess = yield* Effect.tryPromise({
+        try: () => open(url),
+        catch: (error) => (error instanceof Error ? error : new Error(String(error))),
+      })
+      yield* Effect.callback<void, Error>((resume) => {
+        const timer = setTimeout(() => resume(Effect.void), 500)
+        subprocess.on("error", (error) => {
+          clearTimeout(timer)
+          resume(Effect.fail(error))
+        })
+        subprocess.on("exit", (code) => {
+          if (code === null || code === 0) return
+          clearTimeout(timer)
+          resume(Effect.fail(new Error(`Browser open failed with exit code ${code}`)))
+        })
+      })
+    }),
+  }),
+)
+
+export const node = LayerNode.make({ service: Service, layer, deps: [] })
+
+export * as McpBrowser from "./browser"

packages/opencode/src/mcp/index.ts

@@ -1,7 +1,6 @@
 import path from "node:path"
 import { pathToFileURL } from "node:url"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
-import { type Tool } from "ai"
 import { ConfigV1 } from "@opencode-ai/core/v1/config/config"
 import { serviceUse } from "@opencode-ai/core/effect/service-use"
 import { Client, type ClientOptions } from "@modelcontextprotocol/sdk/client/index.js"
@@ -27,7 +26,6 @@ import { McpOAuthCallback } from "./oauth-callback"
 import { McpAuth } from "./auth"
 import { EventV2Bridge } from "@/event-v2-bridge"
 import { TuiEvent } from "@/server/tui-event"
-import open from "open"
 import { Cause, Effect, Exit, Layer, Context, Schema, Stream } from "effect"
 import { EffectBridge } from "@/effect/bridge"
 import { InstanceState } from "@/effect/instance-state"
@@ -35,6 +33,7 @@ import { ChildProcess, ChildProcessSpawner } from "effect/unstable/process"
 import { CrossSpawnSpawner } from "@opencode-ai/core/cross-spawn-spawner"
 import { McpCatalog } from "./catalog"
 import { McpEvent } from "@opencode-ai/schema/mcp-event"
+import { McpBrowser } from "./browser"
 
 const DEFAULT_TIMEOUT = 30_000
 const CLIENT_OPTIONS = {
@@ -154,11 +153,19 @@ export interface ServerInstructions {
   tools: string[]
 }
 
+/** An MCP tool in its native shape; consumers adapt it to their own tool format. */
+export interface McpTool {
+  /** Shared cached definition; consumers must copy rather than mutate it. */
+  readonly def: MCPToolDef
+  readonly client: MCPClient
+  readonly timeout?: number
+}
+
 export interface Interface {
   readonly status: () => Effect.Effect<Record<string, Status>>
   readonly clients: () => Effect.Effect<Record<string, MCPClient>>
   readonly instructions: () => Effect.Effect<ServerInstructions[]>
-  readonly tools: () => Effect.Effect<Record<string, Tool>>
+  readonly tools: () => Effect.Effect<Record<string, McpTool>>
   readonly prompts: () => Effect.Effect<Record<string, PromptInfo & { client: string }>>
   readonly resources: (clientName?: string) => Effect.Effect<Record<string, ResourceInfo & { client: string }>>
   readonly resourceTemplates: (
@@ -200,6 +207,7 @@ const layer = Layer.effect(
     const spawner = yield* ChildProcessSpawner.ChildProcessSpawner
     const auth = yield* McpAuth.Service
     const events = yield* EventV2Bridge.Service
+    const browser = yield* McpBrowser.Service
 
     type Transport = StdioClientTransport | StreamableHTTPClientTransport | SSEClientTransport
 
@@ -656,7 +664,7 @@ const layer = Layer.effect(
     }
 
     const tools = Effect.fn("MCP.tools")(function* () {
-      const result: Record<string, Tool> = {}
+      const result: Record<string, McpTool> = {}
       const s = yield* InstanceState.get(state)
 
       const cfg = yield* cfgSvc.get()
@@ -672,9 +680,8 @@ const layer = Layer.effect(
           continue
         }
         const timeout = requestTimeout(s, clientName, mcpConfig, defaultTimeout)
-        for (const mcpTool of listed) {
-          const key = McpCatalog.toolName(clientName, mcpTool.name)
-          result[key] = McpCatalog.convertTool(mcpTool, client, timeout)
+        for (const def of listed) {
+          result[McpCatalog.toolName(clientName, def.name)] = { def, client, timeout }
         }
       }
       return result
@@ -891,22 +898,7 @@ const layer = Layer.effect(
       const callbackPromise = McpOAuthCallback.waitForCallback(result.oauthState, mcpName)
       onAuthorization?.(result.authorizationUrl)
 
-      yield* Effect.tryPromise(() => open(result.authorizationUrl)).pipe(
-        Effect.flatMap((subprocess) =>
-          Effect.callback<void, Error>((resume) => {
-            const timer = setTimeout(() => resume(Effect.void), 500)
-            subprocess.on("error", (err) => {
-              clearTimeout(timer)
-              resume(Effect.fail(err))
-            })
-            subprocess.on("exit", (code) => {
-              if (code !== null && code !== 0) {
-                clearTimeout(timer)
-                resume(Effect.fail(new Error(`Browser open failed with exit code ${code}`)))
-              }
-            })
-          }),
-        ),
+      yield* browser.open(result.authorizationUrl).pipe(
         Effect.catch(() => {
           return events.publish(BrowserOpenFailed, { mcpName, url: result.authorizationUrl }).pipe(Effect.ignore)
         }),
@@ -1006,7 +998,7 @@ export type AuthStatus = "authenticated" | "expired" | "not_authenticated"
 export const node = LayerNode.make({
   service: Service,
   layer: layer,
-  deps: [CrossSpawnSpawner.node, McpAuth.node, EventV2Bridge.node, Config.node],
+  deps: [CrossSpawnSpawner.node, McpAuth.node, EventV2Bridge.node, Config.node, McpBrowser.node],
 })
 
 export * as MCP from "."

packages/opencode/src/permission/index.ts

@@ -213,6 +213,11 @@ export function disabled(tools: string[], ruleset: PermissionV1.Ruleset): Set<st
   )
 }
 
+export function visibleTools<T>(tools: Record<string, T>, ruleset: PermissionV1.Ruleset): Record<string, T> {
+  const hidden = disabled(Object.keys(tools), ruleset)
+  return Object.fromEntries(Object.entries(tools).filter(([name]) => !hidden.has(name)))
+}
+
 export const node = LayerNode.make({ service: Service, layer: layer, deps: [EventV2Bridge.node] })
 
 export * as Permission from "."

packages/opencode/src/plugin/github-copilot/models.ts

@@ -72,6 +72,10 @@ type SelectableItem = Item & {
     }
   }
 }
+type CopilotEndpoint = "chat" | "responses" | "messages"
+type CopilotModel = Omit<Model, "api"> & {
+  api: Model["api"] & { endpoint?: CopilotEndpoint }
+}
 const decodeModels = Schema.decodeUnknownSync(schema)
 const decodeItem = Schema.decodeUnknownOption(item)
 
@@ -86,17 +90,25 @@ function build(key: string, remote: SelectableItem, url: string, prev?: Model):
     (remote.capabilities.limits.vision?.supported_media_types ?? []).some((item) => item.startsWith("image/"))
 
   const isMsgApi = remote.supported_endpoints?.includes("/v1/messages")
+  const endpoint: CopilotEndpoint | undefined = isMsgApi
+    ? "messages"
+    : remote.supported_endpoints?.includes("/responses")
+      ? "responses"
+      : remote.supported_endpoints?.includes("/chat/completions")
+        ? "chat"
+        : undefined
   const prices = remote.billing?.token_prices
   // Copilot prices are AIC per billing batch; OpenCode stores USD per million tokens.
   const usdPerMillion = prices ? 10_000 / prices.batch_size : 0
 
-  const model: Model = {
+  const model: CopilotModel = {
     id: key,
     providerID: "github-copilot",
     api: {
       id: remote.id,
       url: isMsgApi ? `${url}/v1` : url,
       npm: isMsgApi ? "@ai-sdk/anthropic" : "@ai-sdk/github-copilot",
+      ...(endpoint ? { endpoint } : {}),
     },
     // API response wins
     status: "active",

packages/opencode/src/provider/provider.ts

@@ -218,8 +218,12 @@ function custom(dep: CustomDep): Record<string, CustomLoader> {
     "github-copilot": () =>
       Effect.succeed({
         autoload: false,
-        async getModel(sdk: any, modelID: string, _options?: Record<string, any>) {
+        async getModel(sdk: any, modelID: string, _options?: Record<string, any>, model?: Model) {
           if (sdk.responses === undefined && sdk.chat === undefined) return sdk.languageModel(modelID)
+          if (model && "endpoint" in model.api) {
+            if (model.api.endpoint === "responses" && sdk.responses) return sdk.responses(modelID)
+            if (model.api.endpoint === "chat" && sdk.chat) return sdk.chat(modelID)
+          }
           const match = /^gpt-(\d+)/.exec(modelID)
           if (match && Number(match[1]) >= 5 && !modelID.startsWith("gpt-5-mini")) return sdk.responses(modelID)
           return sdk.chat(modelID)

packages/opencode/src/provider/transform.ts

@@ -1240,9 +1240,6 @@ export function smallOptions(model: Provider.Model) {
     return mergeDeep(base, small)
   }
   if (model.providerID === "openrouter" || model.providerID === "llmgateway") {
-    if (model.providerID === "openrouter" && small.reasoning?.effort === "low") {
-      return { reasoning: { effort: "none" } }
-    }
     if (Object.keys(small).length === 0 && model.api.id.includes("google")) {
       return { reasoning: { enabled: false } }
     }

packages/opencode/src/server/routes/instance/httpapi/handlers/experimental.ts

@@ -137,8 +137,9 @@ export const experimentalHandlers = HttpApiBuilder.group(InstanceHttpApi, "exper
 
     const session = Effect.fn("ExperimentalHttpApi.session")(function* (ctx: { query: typeof SessionListQuery.Type }) {
       const limit = ctx.query.limit ?? 100
+      const directory = ctx.query.directory ? yield* InstanceState.directory : undefined
       const all = yield* sessions.listGlobal({
-        directory: ctx.query.directory,
+        directory,
         roots: ctx.query.roots,
         start: ctx.query.start,
         cursor: ctx.query.cursor,

packages/opencode/src/server/routes/instance/httpapi/handlers/session.ts

@@ -18,6 +18,7 @@ import { MessageID, PartID, SessionID } from "@/session/schema"
 import { NamedError } from "@opencode-ai/core/util/error"
 import { Cause, Effect, Option, Schema, Scope } from "effect"
 import * as Stream from "effect/Stream"
+import { InstanceState } from "@/effect/instance-state"
 import { HttpServerRequest, HttpServerResponse } from "effect/unstable/http"
 import { HttpApiBuilder, HttpApiError, HttpApiSchema } from "effect/unstable/httpapi"
 import { InstanceHttpApi } from "../api"
@@ -61,8 +62,9 @@ export const sessionHandlers = HttpApiBuilder.group(InstanceHttpApi, "session",
     const scope = yield* Scope.Scope
 
     const list = Effect.fn("SessionHttpApi.list")(function* (ctx: { query: typeof ListQuery.Type }) {
+      const directory = ctx.query.directory ? yield* InstanceState.directory : undefined
       return yield* session.list({
-        directory: ctx.query.scope === "project" ? undefined : ctx.query.directory,
+        directory: ctx.query.scope === "project" ? undefined : directory,
         scope: ctx.query.scope,
         path: ctx.query.path,
         roots: ctx.query.roots,

packages/opencode/src/session/processor.ts

@@ -192,6 +192,8 @@ const layer = Layer.effect(
             status: "error",
             input: match.part.state.input,
             error: errorMessage(error),
+            // Keep metadata streamed while running so failures retain progress detail (e.g. execute's child calls).
+            metadata: match.part.state.metadata,
             time: { start: match.part.state.time.start, end: Date.now() },
           },
         })

packages/opencode/src/session/prompt.ts

@@ -1237,6 +1237,7 @@ const layer = Layer.effect(
               Effect.provideService(ToolRegistry.Service, registry),
               Effect.provideService(MCP.Service, mcp),
               Effect.provideService(Truncate.Service, truncate),
+              Effect.provideService(RuntimeFlags.Service, flags),
             )
 
             if (lastUser.format?.type === "json_schema") {

packages/opencode/src/session/tools.ts

@@ -3,6 +3,7 @@ import { SessionV1 } from "@opencode-ai/core/v1/session"
 import { Provider } from "@/provider/provider"
 import { ProviderTransform } from "@/provider/transform"
 import { MCP } from "@/mcp"
+import { McpCatalog } from "@/mcp/catalog"
 import { Permission } from "@/permission"
 import { Tool } from "@/tool/tool"
 import { ToolJsonSchema } from "@/tool/json-schema"
@@ -21,6 +22,7 @@ import { EffectBridge } from "@/effect/bridge"
 import { ProviderV2 } from "@opencode-ai/core/provider"
 import { ModelV2 } from "@opencode-ai/core/model"
 import { isRecord } from "@/util/record"
+import { RuntimeFlags } from "@/effect/runtime-flags"
 
 const MCP_RESOURCE_TOOLS = {
   list: "list_mcp_resources",
@@ -52,6 +54,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
   const registry = yield* ToolRegistry.Service
   const mcp = yield* MCP.Service
   const truncate = yield* Truncate.Service
+  const flags = yield* RuntimeFlags.Service
 
   const context = (args: Record<string, unknown>, options: ToolExecutionOptions): Tool.Context => ({
     sessionID: input.session.id,
@@ -90,6 +93,7 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
     modelID: ModelV2.ID.make(input.model.api.id),
     providerID: input.model.providerID,
     agent: input.agent,
+    permission: input.session.permission,
   })) {
     const schema = ProviderTransform.schema(input.model, ToolJsonSchema.fromTool(item))
     tools[item.id] = tool({
@@ -381,7 +385,10 @@ export const resolve = Effect.fn("SessionTools.resolve")(function* (input: {
     })
   }
 
-  for (const [key, item] of Object.entries(yield* mcp.tools())) {
+  if (flags.experimentalCodeMode) return tools
+
+  for (const [key, entry] of Object.entries(yield* mcp.tools())) {
+    const item = McpCatalog.convertTool(entry.def, entry.client, entry.timeout)
     const execute = item.execute
     if (!execute) continue
 

packages/opencode/src/tool/code-mode.ts

@@ -0,0 +1,310 @@
+import * as Tool from "./tool"
+import { CallToolResultSchema, type CallToolResult } from "@modelcontextprotocol/sdk/types.js"
+import { Cause, Effect, Schema } from "effect"
+import { CodeMode, Tool as SandboxTool, toolError } from "@opencode-ai/codemode"
+import { MCP } from "@/mcp"
+import { McpCatalog } from "@/mcp/catalog"
+import { Agent } from "@/agent/agent"
+import { Session } from "@/session/session"
+import { Permission } from "@/permission"
+import { Plugin } from "@/plugin"
+
+export const CODE_MODE_TOOL = "execute"
+
+const DESCRIPTION = "Run a confined orchestration script with access to connected MCP tools."
+
+export const Parameters = Schema.Struct({
+  code: Schema.String.annotate({
+    description: "Script body executed by the confined interpreter.",
+  }),
+})
+
+type CallEntry = { tool: string; status: "running" | "completed" | "error"; input?: Record<string, unknown> }
+
+type Metadata = {
+  toolCalls: CallEntry[]
+  error?: boolean
+}
+
+type Attachment = NonNullable<Tool.ExecuteResult["attachments"]>[number]
+
+type CatalogEntry = {
+  path: string
+  key: string
+  server: string
+  local: string
+  tool: MCP.McpTool
+}
+
+function groupByServer(mcpTools: Record<string, MCP.McpTool>, servers: readonly string[]): Map<string, CatalogEntry[]> {
+  const byLongest = [...servers].sort((a, b) => b.length - a.length)
+  const groups = new Map<string, CatalogEntry[]>()
+  for (const key of Object.keys(mcpTools).sort((a, b) => a.localeCompare(b))) {
+    const server =
+      byLongest.find((name) => key.startsWith(name + "_")) ?? (key.includes("_") ? key.slice(0, key.indexOf("_")) : key)
+    const local = server && key.startsWith(server + "_") ? key.slice(server.length + 1) : key
+    const entry: CatalogEntry = {
+      path: `${server}.${local}`,
+      key,
+      server,
+      local,
+      tool: mcpTools[key]!,
+    }
+    groups.set(server, [...(groups.get(server) ?? []), entry])
+  }
+  return groups
+}
+
+export function describeCatalog(mcpTools: Record<string, MCP.McpTool>, servers: readonly string[]): string {
+  return CodeMode.make({
+    tools: toolTree(
+      [...groupByServer(mcpTools, servers).values()].flat(),
+      () => () => Effect.fail(toolError("Tool preview is not executable.")),
+    ),
+  }).instructions()
+}
+
+const lastSegment = (uri: string) => {
+  const trimmed = uri.split(/[?#]/, 1)[0]!.replace(/\/+$/, "")
+  const segment = trimmed.slice(trimmed.lastIndexOf("/") + 1)
+  return segment.length > 0 ? segment : undefined
+}
+
+const dataUrl = (mime: string, base64: string) => `data:${mime};base64,${base64}`
+
+function projectMcpResult(result: CallToolResult, collect: (attachment: Attachment) => void): unknown {
+  const text: string[] = []
+  let files = 0
+  let images = 0
+  const push = (attachment: Attachment) => {
+    files += 1
+    if (attachment.mime.startsWith("image/")) images += 1
+    collect(attachment)
+  }
+  for (const block of result.content) {
+    switch (block.type) {
+      case "text":
+        text.push(block.text)
+        break
+      case "image":
+      case "audio":
+        push({ type: "file", mime: block.mimeType, url: dataUrl(block.mimeType, block.data) })
+        break
+      case "resource": {
+        if ("text" in block.resource) {
+          text.push(block.resource.text)
+          break
+        }
+        const mime = block.resource.mimeType ?? "application/octet-stream"
+        push({ type: "file", mime, url: dataUrl(mime, block.resource.blob), filename: lastSegment(block.resource.uri) })
+        break
+      }
+      case "resource_link":
+        // A link is a reference, not fetchable media; hand it to the program instead of the attachment channel.
+        text.push(`${block.name}: ${block.uri}`)
+        break
+    }
+  }
+
+  if (result.structuredContent !== undefined && result.structuredContent !== null) return result.structuredContent
+  if (text.length > 0) return text.join("\n")
+  if (files > 0) {
+    const noun = files === images ? "image" : "file"
+    return `[${files} ${noun}${files === 1 ? "" : "s"} attached to the result]`
+  }
+  return null
+}
+
+type Run = (input: unknown) => Effect.Effect<unknown, unknown>
+
+function toolTree(catalog: readonly CatalogEntry[], run: (entry: CatalogEntry) => Run) {
+  const tree: Record<string, Record<string, SandboxTool.Definition>> = {}
+  for (const entry of catalog) {
+    const namespace = (tree[entry.server] ??= {})
+    namespace[entry.local] = SandboxTool.make({
+      description: entry.tool.def.description ?? "",
+      input: entry.tool.def.inputSchema as SandboxTool.JsonSchema,
+      output: entry.tool.def.outputSchema as SandboxTool.JsonSchema | undefined,
+      run: run(entry),
+    })
+  }
+  return tree
+}
+
+const invokeChildTool = Effect.fn("CodeMode.invokeChildTool")(function* (input: {
+  plugin: Plugin.Interface
+  entry: CatalogEntry
+  args: Record<string, unknown>
+  callID: string
+  ctx: Tool.Context
+}) {
+  yield* input.plugin.trigger(
+    "tool.execute.before",
+    { tool: input.entry.key, sessionID: input.ctx.sessionID, callID: input.callID },
+    { args: input.args },
+  )
+  const result: CallToolResult = yield* Effect.gen(function* () {
+    yield* input.ctx.ask({ permission: input.entry.key, metadata: {}, patterns: ["*"], always: ["*"] })
+    // Deliberately mirrors McpCatalog.convertTool's transport call so the MCP service stays free of tool-loop concerns.
+    return yield* Effect.promise(async () => {
+      const raw = await input.entry.tool.client.callTool(
+        { name: input.entry.tool.def.name, arguments: input.args },
+        CallToolResultSchema,
+        {
+          resetTimeoutOnProgress: true,
+          signal: input.ctx.abort,
+          timeout: input.entry.tool.timeout,
+          // The MCP SDK only sends a progress token when this hook is present, enabling timeout resets.
+          onprogress: () => {},
+        },
+      )
+      if (raw.isError)
+        throw new Error(
+          raw.content
+            .flatMap((item) => (item.type === "text" ? [item.text] : []))
+            .filter((text) => text.trim())
+            .join("\n\n") || "MCP tool returned an error",
+        )
+      return raw
+    })
+  }).pipe(
+    Effect.withSpan("Tool.execute", {
+      attributes: {
+        "tool.name": input.entry.key,
+        "tool.call_id": input.callID,
+        "session.id": input.ctx.sessionID,
+        "message.id": input.ctx.messageID,
+      },
+    }),
+  )
+  yield* input.plugin.trigger(
+    "tool.execute.after",
+    { tool: input.entry.key, sessionID: input.ctx.sessionID, callID: input.callID, args: input.args },
+    result,
+  )
+  return result
+})
+
+export const CodeModeTool = Tool.define(
+  CODE_MODE_TOOL,
+  Effect.gen(function* () {
+    const mcp = yield* MCP.Service
+    const agents = yield* Agent.Service
+    const sessions = yield* Session.Service
+    const plugin = yield* Plugin.Service
+
+    const init: Tool.DefWithoutID<typeof Parameters, Metadata> = {
+      description: DESCRIPTION,
+      parameters: Parameters,
+      execute: Effect.fn("CodeMode.execute")(function* (params, ctx) {
+        if (ctx.abort.aborted) {
+          return {
+            title: CODE_MODE_TOOL,
+            metadata: { toolCalls: [], error: true },
+            output: "Execution cancelled.",
+          } satisfies Tool.ExecuteResult<Metadata>
+        }
+        const agent = yield* agents.get(ctx.agent)
+        const session = yield* sessions.get(ctx.sessionID).pipe(Effect.orDie)
+        const ruleset = Permission.merge(agent.permission, session.permission ?? [])
+        const mcpTools = Permission.visibleTools(yield* mcp.tools(), ruleset)
+        const servers = Object.keys(yield* mcp.clients()).map(McpCatalog.sanitize)
+        const catalog = [...groupByServer(mcpTools, servers).values()].flat()
+
+        const calls: CallEntry[] = []
+        const attachments: Attachment[] = []
+        const publish = () =>
+          ctx.metadata({ title: CODE_MODE_TOOL, metadata: { toolCalls: calls.map((c) => ({ ...c })) } })
+
+        let childCalls = 0
+        const callTool = (entry: CatalogEntry) => (input: unknown) =>
+          Effect.gen(function* () {
+            childCalls += 1
+            const result = yield* invokeChildTool({
+              plugin,
+              entry,
+              args: (input ?? {}) as Record<string, unknown>,
+              callID: `${ctx.callID ?? entry.key}/${childCalls}`,
+              ctx,
+            })
+            return projectMcpResult(result, (attachment: Attachment) => void attachments.push(attachment))
+          }).pipe(
+            Effect.catchCause((cause) => {
+              if (Cause.hasInterruptsOnly(cause)) return Effect.interrupt
+              const error = Cause.squash(cause)
+              return Effect.fail(toolError(error instanceof Error ? error.message : String(error), error))
+            }),
+          )
+
+        const runtime = CodeMode.make({
+          tools: toolTree(catalog, callTool),
+          onToolCallStart: ({ index, name, input }) =>
+            Effect.suspend(() => {
+              const shown = (() => {
+                if (input === null || input === undefined) return
+                if (typeof input === "object" && !Array.isArray(input)) {
+                  const value = input as Record<string, unknown>
+                  return Object.keys(value).length > 0 ? value : undefined
+                }
+                return { input }
+              })()
+              calls[index] = { tool: name, status: "running", ...(shown ? { input: shown } : {}) }
+              return publish()
+            }),
+          onToolCallEnd: ({ index, outcome }) =>
+            Effect.suspend(() => {
+              const current = calls[index]
+              if (current) calls[index] = { ...current, status: outcome === "success" ? "completed" : "error" }
+              return publish()
+            }),
+        })
+
+        const abort = Effect.callback<void>((resume) => {
+          if (ctx.abort.aborted) return resume(Effect.void)
+          const handler = () => resume(Effect.void)
+          ctx.abort.addEventListener("abort", handler, { once: true })
+          return Effect.sync(() => ctx.abort.removeEventListener("abort", handler))
+        })
+        const cancelled = (): CodeMode.Result => ({
+          ok: false,
+          error: { kind: "ExecutionFailure", message: "Execution cancelled." },
+          toolCalls: calls.map((call) => ({ name: call.tool })),
+        })
+
+        const result = yield* Effect.raceFirst(runtime.execute(params.code), abort.pipe(Effect.map(cancelled)))
+        const logs = result.logs ?? []
+        const withLogs = (text: string) => {
+          if (logs.length === 0) return text
+          return text.length > 0 ? `${text}\n\nLogs:\n${logs.join("\n")}` : `Logs:\n${logs.join("\n")}`
+        }
+
+        if (!result.ok) {
+          if (ctx.abort.aborted) {
+            return {
+              title: CODE_MODE_TOOL,
+              metadata: { toolCalls: calls, error: true },
+              output: "Execution cancelled.",
+            } satisfies Tool.ExecuteResult<Metadata>
+          }
+          const hints = (result.error.suggestions ?? []).filter((hint) => !result.error.message.includes(hint))
+          return yield* Effect.fail(new Error(withLogs([result.error.message, ...hints].join("\n"))))
+        }
+
+        // The interpreter validates returned values as plain JSON, so stringify cannot throw;
+        // it yields undefined only for a program that returns undefined.
+        const output =
+          typeof result.value === "string"
+            ? result.value
+            : (JSON.stringify(result.value, null, 2) ?? String(result.value))
+
+        return {
+          title: CODE_MODE_TOOL,
+          metadata: { toolCalls: calls },
+          output: withLogs(output),
+          ...(attachments.length > 0 ? { attachments } : {}),
+        } satisfies Tool.ExecuteResult<Metadata>
+      }, Effect.orDie),
+    }
+    return init
+  }),
+)

packages/opencode/src/tool/registry.ts

@@ -51,6 +51,9 @@ import { Job } from "@/job"
 import { RuntimeFlags } from "@/effect/runtime-flags"
 import { ProviderV2 } from "@opencode-ai/core/provider"
 import { ModelV2 } from "@opencode-ai/core/model"
+import { MCP } from "@/mcp"
+import { PermissionV1 } from "@opencode-ai/core/v1/permission"
+import { McpCatalog } from "@/mcp/catalog"
 
 export function webSearchEnabled(providerID: ProviderV2.ID, flags = { exa: false, parallel: false }) {
   return providerID === ProviderV2.ID.opencode || flags.exa || flags.parallel
@@ -74,6 +77,7 @@ export interface Interface {
     providerID: ProviderV2.ID
     modelID: ModelV2.ID
     agent: Agent.Info
+    permission?: PermissionV1.Ruleset
   }) => Effect.Effect<Tool.Def[]>
 }
 
@@ -87,6 +91,7 @@ const layer = Layer.effect(
     const agents = yield* Agent.Service
     const truncate = yield* Truncate.Service
     const flags = yield* RuntimeFlags.Service
+    const mcp = yield* MCP.Service
 
     const invalid = yield* InvalidTool
     const task = yield* TaskTool
@@ -105,6 +110,8 @@ const layer = Layer.effect(
     const patchtool = yield* ApplyPatchTool
     const skilltool = yield* SkillTool
     const agent = yield* Agent.Service
+    const codeMode = flags.experimentalCodeMode ? yield* Effect.promise(() => import("./code-mode")) : undefined
+    const codeModeTool = codeMode ? yield* codeMode.CodeModeTool : undefined
 
     const state = yield* InstanceState.make<State>(
       Effect.fn("ToolRegistry.state")(function* (ctx) {
@@ -211,6 +218,7 @@ const layer = Layer.effect(
           question: Tool.init(question),
           lsp: Tool.init(lsptool),
           plan: Tool.init(plan),
+          ...(codeModeTool ? { execute: Tool.init(codeModeTool) } : {}),
         })
 
         return {
@@ -230,6 +238,7 @@ const layer = Layer.effect(
             tool.search,
             tool.skill,
             tool.patch,
+            ...(tool.execute ? [tool.execute] : []),
             ...(flags.experimentalLspTool ? [tool.lsp] : []),
             ...(flags.experimentalPlanMode && flags.client === "cli" ? [tool.plan] : []),
           ],
@@ -263,6 +272,17 @@ const layer = Layer.effect(
       return ["Available agent types and the tools they have access to:", description].join("\n")
     })
 
+    const describeCodeMode = Effect.fn("ToolRegistry.describeCodeMode")(function* (input: {
+      agent: Agent.Info
+      permission?: PermissionV1.Ruleset
+    }) {
+      if (!codeMode) return
+      const ruleset = Permission.merge(input.agent.permission, input.permission ?? [])
+      const tools = Permission.visibleTools(yield* mcp.tools(), ruleset)
+      if (Object.keys(tools).length === 0) return
+      return codeMode.describeCatalog(tools, Object.keys(yield* mcp.clients()).map(McpCatalog.sanitize))
+    })
+
     const tools: Interface["tools"] = Effect.fn("ToolRegistry.tools")(function* (input) {
       const filtered = (yield* all()).filter((tool) => {
         if (tool.id === WebSearchTool.id) {
@@ -277,8 +297,13 @@ const layer = Layer.effect(
         return true
       })
 
+      const codeModeDescription = filtered.some((tool) => tool.id === "execute")
+        ? yield* describeCodeMode(input)
+        : undefined
+      const visible = filtered.filter((tool) => tool.id !== "execute" || codeModeDescription)
+
       return yield* Effect.forEach(
-        filtered,
+        visible,
         Effect.fnUntraced(function* (tool: Tool.Def) {
           const output = {
             description: tool.description,
@@ -292,7 +317,11 @@ const layer = Layer.effect(
               : undefined
           return {
             id: tool.id,
-            description: [output.description, tool.id === TaskTool.id ? yield* describeTask(input.agent) : undefined]
+            description: [
+              output.description,
+              tool.id === TaskTool.id ? yield* describeTask(input.agent) : undefined,
+              tool.id === "execute" ? codeModeDescription : undefined,
+            ]
               .filter(Boolean)
               .join("\n"),
             parameters: output.parameters,
@@ -412,6 +441,7 @@ export const node = LayerNode.make({
     Format.node,
     Truncate.node,
     RuntimeFlags.node,
+    MCP.node,
     Database.node,
     Ripgrep.node,
   ],

packages/opencode/test/fixture/mcp-lifecycle-stdio.ts

@@ -0,0 +1,26 @@
+import { Server } from "@modelcontextprotocol/sdk/server/index.js"
+import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js"
+import { ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js"
+
+if (process.argv.includes("--hang")) {
+  const pidFile = process.env.MCP_LIFECYCLE_PID_FILE
+  if (!pidFile) throw new Error("MCP_LIFECYCLE_PID_FILE is required")
+  await Bun.write(pidFile, String(process.pid))
+  await new Promise(() => {})
+}
+
+const server = new Server({ name: "mcp-lifecycle-stdio", version: "1.0.0" }, { capabilities: { tools: {} } })
+
+server.setRequestHandler(ListToolsRequestSchema, () =>
+  Promise.resolve({
+    tools: [
+      {
+        name: "current_directory",
+        description: process.cwd(),
+        inputSchema: { type: "object", properties: {} },
+      },
+    ],
+  }),
+)
+
+await server.connect(new StdioServerTransport())

packages/opencode/test/mcp/headers.test.ts

@@ -1,126 +1,101 @@
-import { describe, expect, mock, beforeEach } from "bun:test"
+import { describe, expect } from "bun:test"
+import { Server } from "@modelcontextprotocol/sdk/server/index.js"
+import { WebStandardStreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/webStandardStreamableHttp.js"
+import { ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
 import { Effect } from "effect"
 import { testEffect } from "../lib/effect"
+import { MCP } from "../../src/mcp/index"
 
-// Track what options were passed to each transport constructor
-const transportCalls: Array<{
-  type: "streamable" | "sse"
-  url: string
-  options: { authProvider?: unknown; requestInit?: RequestInit }
-}> = []
-
-// Mock the transport constructors to capture their arguments
-void mock.module("@modelcontextprotocol/sdk/client/streamableHttp.js", () => ({
-  StreamableHTTPClientTransport: class MockStreamableHTTP {
-    constructor(url: URL, options?: { authProvider?: unknown; requestInit?: RequestInit }) {
-      transportCalls.push({
-        type: "streamable",
-        url: url.toString(),
-        options: options ?? {},
-      })
-    }
-    async start() {
-      throw new Error("Mock transport cannot connect")
-    }
-  },
-}))
-
-void mock.module("@modelcontextprotocol/sdk/client/sse.js", () => ({
-  SSEClientTransport: class MockSSE {
-    constructor(url: URL, options?: { authProvider?: unknown; requestInit?: RequestInit }) {
-      transportCalls.push({
-        type: "sse",
-        url: url.toString(),
-        options: options ?? {},
-      })
-    }
-    async start() {
-      throw new Error("Mock transport cannot connect")
-    }
-  },
-}))
-
-beforeEach(() => {
-  transportCalls.length = 0
-})
-
-// Import MCP after mocking
-const { MCP } = await import("../../src/mcp/index")
 const it = testEffect(LayerNode.compile(MCP.node))
 
+const serve = Effect.acquireRelease(
+  Effect.promise(async () => {
+    const requests: Headers[] = []
+    const protocol = new Server({ name: "headers", version: "1.0.0" }, { capabilities: { tools: {} } })
+    protocol.setRequestHandler(ListToolsRequestSchema, () => Promise.resolve({ tools: [] }))
+    const transport = new WebStandardStreamableHTTPServerTransport({
+      sessionIdGenerator: () => crypto.randomUUID(),
+      enableJsonResponse: true,
+    })
+    await protocol.connect(transport)
+    const http = Bun.serve({
+      port: 0,
+      fetch(request) {
+        requests.push(new Headers(request.headers))
+        return transport.handleRequest(request)
+      },
+    })
+    return {
+      requests,
+      url: http.url.toString(),
+      close: async () => {
+        await http.stop(true)
+        await protocol.close()
+      },
+    }
+  }),
+  (server) => Effect.promise(server.close),
+)
+
 describe("mcp.headers", () => {
   it.instance("headers are passed to transports when oauth is enabled (default)", () =>
     Effect.gen(function* () {
+      const server = yield* serve
       const mcp = yield* MCP.Service
-      yield* mcp
-        .add("test-server", {
-          type: "remote",
-          url: "https://example.com/mcp",
-          headers: {
-            Authorization: "Bearer test-token",
-            "X-Custom-Header": "custom-value",
-          },
-        })
-        .pipe(Effect.catch(() => Effect.void))
-
-      // Both transports should have been created with headers
-      expect(transportCalls.length).toBeGreaterThanOrEqual(1)
-
-      for (const call of transportCalls) {
-        expect(call.options.requestInit).toBeDefined()
-        expect(call.options.requestInit?.headers).toEqual({
+      const result = yield* mcp.add("test-server", {
+        type: "remote",
+        url: server.url,
+        headers: {
           Authorization: "Bearer test-token",
           "X-Custom-Header": "custom-value",
-        })
-        // OAuth should be enabled by default, so authProvider should exist
-        expect(call.options.authProvider).toBeDefined()
+        },
+      })
+
+      expect(result.status).toMatchObject({ "test-server": { status: "connected" } })
+      expect(server.requests.length).toBeGreaterThan(0)
+      for (const headers of server.requests) {
+        expect(headers.get("authorization")).toBe("Bearer test-token")
+        expect(headers.get("x-custom-header")).toBe("custom-value")
       }
     }),
   )
 
   it.instance("headers are passed to transports when oauth is explicitly disabled", () =>
     Effect.gen(function* () {
+      const server = yield* serve
       const mcp = yield* MCP.Service
-      yield* mcp
-        .add("test-server-no-oauth", {
-          type: "remote",
-          url: "https://example.com/mcp",
-          oauth: false,
-          headers: {
-            Authorization: "Bearer test-token",
-          },
-        })
-        .pipe(Effect.catch(() => Effect.void))
-
-      expect(transportCalls.length).toBeGreaterThanOrEqual(1)
-
-      for (const call of transportCalls) {
-        expect(call.options.requestInit).toBeDefined()
-        expect(call.options.requestInit?.headers).toEqual({
+      const result = yield* mcp.add("test-server-no-oauth", {
+        type: "remote",
+        url: server.url,
+        oauth: false,
+        headers: {
           Authorization: "Bearer test-token",
-        })
-        // OAuth is disabled, so no authProvider
-        expect(call.options.authProvider).toBeUndefined()
+        },
+      })
+
+      expect(result.status).toMatchObject({ "test-server-no-oauth": { status: "connected" } })
+      expect(server.requests.length).toBeGreaterThan(0)
+      for (const headers of server.requests) {
+        expect(headers.get("authorization")).toBe("Bearer test-token")
       }
     }),
   )
 
   it.instance("no requestInit when headers are not provided", () =>
     Effect.gen(function* () {
+      const server = yield* serve
       const mcp = yield* MCP.Service
-      yield* mcp
-        .add("test-server-no-headers", {
-          type: "remote",
-          url: "https://example.com/mcp",
-        })
-        .pipe(Effect.catch(() => Effect.void))
+      const result = yield* mcp.add("test-server-no-headers", {
+        type: "remote",
+        url: server.url,
+      })
 
-      expect(transportCalls.length).toBeGreaterThanOrEqual(1)
-
-      for (const call of transportCalls) {
-        // No headers means requestInit should be undefined
-        expect(call.options.requestInit).toBeUndefined()
+      expect(result.status).toMatchObject({ "test-server-no-headers": { status: "connected" } })
+      expect(server.requests.length).toBeGreaterThan(0)
+      for (const headers of server.requests) {
+        expect(headers.has("authorization")).toBe(false)
+        expect(headers.has("x-custom-header")).toBe(false)
       }
     }),
   )

packages/opencode/test/mcp/oauth-auto-connect.test.ts

@@ -1,199 +1,155 @@
-import { expect, mock, beforeEach } from "bun:test"
+import { expect } from "bun:test"
+import { Server } from "@modelcontextprotocol/sdk/server/index.js"
+import { WebStandardStreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/webStandardStreamableHttp.js"
+import { ListResourcesRequestSchema, ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
-import { Effect, Layer } from "effect"
+import { CrossSpawnSpawner } from "@opencode-ai/core/cross-spawn-spawner"
+import { FSUtil } from "@opencode-ai/core/fs-util"
+import { Effect } from "effect"
+import { Config } from "../../src/config/config"
+import { EventV2Bridge } from "../../src/event-v2-bridge"
+import { McpAuth } from "../../src/mcp/auth"
+import { MCP } from "../../src/mcp/index"
+import { McpOAuthCallback } from "../../src/mcp/oauth-callback"
+import { McpOAuthPendingProvider, McpOAuthProvider } from "../../src/mcp/oauth-provider"
 import { testEffect } from "../lib/effect"
 
-// Mock UnauthorizedError to match the SDK's class
-class MockUnauthorizedError extends Error {
-  constructor(message?: string) {
-    super(message ?? "Unauthorized")
-    this.name = "UnauthorizedError"
-  }
-}
-
-// Track what options were passed to each transport constructor
-const transportCalls: Array<{
-  type: "streamable" | "sse"
-  url: string
-  options: { authProvider?: unknown }
-}> = []
-
-// Controls whether the mock transport simulates a 401 that triggers the SDK
-// auth flow (which calls provider.state()) or a simple UnauthorizedError.
-let simulateAuthFlow = true
-let connectSucceedsImmediately = false
-let serverCapabilities: { tools?: object; resources?: object } = { tools: {} }
-let listToolsCalls = 0
-let finishAuthFails = false
-let finishAuthStoresCredentials = false
-
-// Mock the transport constructors to simulate OAuth auto-auth on 401
-void mock.module("@modelcontextprotocol/sdk/client/streamableHttp.js", () => ({
-  StreamableHTTPClientTransport: class MockStreamableHTTP {
-    authProvider:
-      | {
-          state?: () => Promise<string>
-          redirectToAuthorization?: (url: URL) => Promise<void>
-          saveCodeVerifier?: (v: string) => Promise<void>
-          tokens?: () => Promise<{ access_token: string } | undefined>
-          clientInformation?: () => Promise<{ client_id: string } | undefined>
-          saveClientInformation?: (info: { client_id: string; client_secret?: string }) => Promise<void>
-          saveTokens?: (tokens: { access_token: string; token_type: string }) => Promise<void>
-        }
-      | undefined
-    constructor(url: URL, options?: { authProvider?: unknown }) {
-      this.authProvider = options?.authProvider as typeof this.authProvider
-      transportCalls.push({
-        type: "streamable",
-        url: url.toString(),
-        options: options ?? {},
-      })
-    }
-    async start() {
-      if (connectSucceedsImmediately) return
-
-      // Simulate what the real SDK transport does on 401:
-      // It calls auth() which eventually calls provider.state(), then
-      // provider.redirectToAuthorization(), then throws UnauthorizedError.
-      if (simulateAuthFlow && this.authProvider) {
-        if (await this.authProvider.tokens?.()) throw new MockUnauthorizedError()
-        if (await this.authProvider.clientInformation?.()) throw new MockUnauthorizedError()
-        // The SDK calls provider.state() to get the OAuth state parameter
-        if (this.authProvider.state) {
-          await this.authProvider.state()
-        }
-        // The SDK calls saveCodeVerifier before redirecting
-        if (this.authProvider.saveCodeVerifier) {
-          await this.authProvider.saveCodeVerifier("test-verifier")
-        }
-        // The SDK calls redirectToAuthorization to redirect the user
-        if (this.authProvider.redirectToAuthorization) {
-          await this.authProvider.redirectToAuthorization(new URL("https://auth.example.com/authorize?state=test"))
-        }
-        throw new MockUnauthorizedError()
-      }
-      throw new MockUnauthorizedError()
-    }
-    async finishAuth(_code: string) {
-      if (finishAuthFails) throw new Error("Token exchange failed")
-      if (finishAuthStoresCredentials) {
-        await this.authProvider?.saveClientInformation?.({ client_id: "replacement-client" })
-        await this.authProvider?.saveTokens?.({ access_token: "replacement-token", token_type: "Bearer" })
-      }
-    }
-    async close() {}
-  },
-}))
-
-void mock.module("@modelcontextprotocol/sdk/client/sse.js", () => ({
-  SSEClientTransport: class MockSSE {
-    constructor(url: URL, options?: { authProvider?: unknown }) {
-      transportCalls.push({
-        type: "sse",
-        url: url.toString(),
-        options: options ?? {},
-      })
-    }
-    async start() {
-      throw new Error("Mock SSE transport cannot connect")
-    }
-  },
-}))
-
-// Mock the MCP SDK Client
-void mock.module("@modelcontextprotocol/sdk/client/index.js", () => ({
-  Client: class MockClient {
-    setRequestHandler() {}
-
-    async connect(transport: { start: () => Promise<void> }) {
-      await transport.start()
-    }
-
-    setNotificationHandler() {}
-
-    getServerCapabilities() {
-      return serverCapabilities
-    }
-
-    getInstructions() {}
-
-    async listTools() {
-      listToolsCalls++
-      return { tools: [{ name: "test_tool", inputSchema: { type: "object", properties: {} } }] }
-    }
-
-    async listResources() {
-      return { resources: [{ name: "docs", uri: "docs://readme" }] }
-    }
-
-    async close() {}
-  },
-}))
-
-// Mock UnauthorizedError in the auth module so instanceof checks work
-void mock.module("@modelcontextprotocol/sdk/client/auth.js", () => ({
-  UnauthorizedError: MockUnauthorizedError,
-}))
-
-beforeEach(() => {
-  transportCalls.length = 0
-  simulateAuthFlow = true
-  connectSucceedsImmediately = false
-  serverCapabilities = { tools: {} }
-  listToolsCalls = 0
-  finishAuthFails = false
-  finishAuthStoresCredentials = false
-})
-
-// Import modules after mocking
-const { MCP } = await import("../../src/mcp/index")
-const { EventV2Bridge } = await import("../../src/event-v2-bridge")
-const { Config } = await import("../../src/config/config")
-const { McpAuth } = await import("../../src/mcp/auth")
-const { McpOAuthProvider } = await import("../../src/mcp/oauth-provider")
-const { McpOAuthCallback } = await import("../../src/mcp/oauth-callback")
-const { FSUtil } = await import("@opencode-ai/core/fs-util")
-const { CrossSpawnSpawner } = await import("@opencode-ai/core/cross-spawn-spawner")
-
 const mcpTest = testEffect(
   LayerNode.compile(
     LayerNode.group([MCP.node, McpAuth.node, EventV2Bridge.node, Config.node, CrossSpawnSpawner.node, FSUtil.node]),
   ),
 )
 
-const config = (name: string) => ({
-  mcp: {
-    [name]: {
-      type: "remote" as const,
-      url: "https://example.com/mcp",
-    },
-  },
+interface OAuthMcpOptions {
+  capabilities?: "tools" | "resources"
+}
+
+function serveOAuthMcp(options: OAuthMcpOptions = {}) {
+  return Effect.acquireRelease(
+    Effect.promise(async () => {
+      const capabilities = options.capabilities ?? "tools"
+      const protocol = new Server(
+        { name: "oauth-auto-connect", version: "1.0.0" },
+        { capabilities: capabilities === "tools" ? { tools: {} } : { resources: {} } },
+      )
+      const transport = new WebStandardStreamableHTTPServerTransport({
+        sessionIdGenerator: () => crypto.randomUUID(),
+        enableJsonResponse: true,
+      })
+      let listToolsCalls = 0
+      let requiresAuth = true
+
+      if (capabilities === "tools") {
+        protocol.setRequestHandler(ListToolsRequestSchema, () => {
+          listToolsCalls++
+          return Promise.resolve({ tools: [{ name: "test_tool", inputSchema: { type: "object" } }] })
+        })
+      }
+      if (capabilities === "resources") {
+        protocol.setRequestHandler(ListResourcesRequestSchema, () =>
+          Promise.resolve({ resources: [{ name: "docs", uri: "docs://readme" }] }),
+        )
+      }
+
+      await protocol.connect(transport)
+      const http = Bun.serve({
+        port: 0,
+        async fetch(request) {
+          const url = new URL(request.url)
+          const origin = url.origin
+          const mcpUrl = `${origin}/mcp`
+
+          if (url.pathname === "/.well-known/oauth-protected-resource/mcp") {
+            return Response.json({
+              resource: mcpUrl,
+              authorization_servers: [origin],
+              scopes_supported: ["mcp"],
+            })
+          }
+          if (url.pathname === "/.well-known/oauth-protected-resource") {
+            return Response.json({
+              resource: mcpUrl,
+              authorization_servers: [origin],
+              scopes_supported: ["mcp"],
+            })
+          }
+          if (url.pathname === "/.well-known/oauth-authorization-server") {
+            return Response.json({
+              issuer: origin,
+              authorization_endpoint: `${origin}/authorize`,
+              token_endpoint: `${origin}/token`,
+              registration_endpoint: `${origin}/register`,
+              response_types_supported: ["code"],
+              grant_types_supported: ["authorization_code", "refresh_token"],
+              token_endpoint_auth_methods_supported: ["none"],
+              code_challenge_methods_supported: ["S256"],
+              scopes_supported: ["mcp"],
+            })
+          }
+          if (url.pathname === "/register") {
+            const metadata = (await request.json()) as Record<string, unknown>
+            return Response.json({ ...metadata, client_id: "replacement-client" }, { status: 201 })
+          }
+          if (url.pathname === "/token") {
+            const body = new URLSearchParams(await request.text())
+            if (body.get("code") !== "valid-code") {
+              return Response.json(
+                { error: "invalid_grant", error_description: "Token exchange failed" },
+                { status: 400 },
+              )
+            }
+            return Response.json({ access_token: "replacement-token", token_type: "Bearer" })
+          }
+          if (url.pathname !== "/mcp") return new Response("Not found", { status: 404 })
+
+          if (request.method === "GET") return new Response(null, { status: 405 })
+          if (requiresAuth && request.headers.get("authorization") !== "Bearer replacement-token") {
+            return new Response("Unauthorized", {
+              status: 401,
+              headers: {
+                "WWW-Authenticate": `Bearer resource_metadata="${origin}/.well-known/oauth-protected-resource", scope="mcp"`,
+              },
+            })
+          }
+          return transport.handleRequest(request)
+        },
+      })
+
+      return {
+        url: new URL("/mcp", http.url).toString(),
+        allowAnonymous: () => {
+          requiresAuth = false
+        },
+        listToolsCalls: () => listToolsCalls,
+        close: async () => {
+          await http.stop(true)
+          await protocol.close()
+        },
+      }
+    }),
+    (server) => Effect.promise(server.close),
+  )
+}
+
+const remote = (url: string, enabled = true) => ({
+  type: "remote" as const,
+  url,
+  enabled,
 })
 
-mcpTest.instance(
-  "first connect to OAuth server shows needs_auth instead of failed",
-  () =>
-    MCP.Service.use((mcp) =>
-      Effect.gen(function* () {
-        const result = yield* mcp.add("test-oauth", {
-          type: "remote",
-          url: "https://example.com/mcp",
-        })
+const stopOAuthCallback = Effect.addFinalizer(() => Effect.promise(() => McpOAuthCallback.stop()).pipe(Effect.ignore))
 
-        const serverStatus = result.status as Record<string, { status: string; error?: string }>
+mcpTest.instance("first connect to OAuth server shows needs_auth instead of failed", () =>
+  Effect.gen(function* () {
+    const server = yield* serveOAuthMcp()
+    const mcp = yield* MCP.Service
+    const result = yield* mcp.add("test-oauth", remote(server.url))
 
-        // The server should be detected as needing auth, NOT as failed.
-        // Before the fix, provider.state() would throw a plain Error
-        // ("No OAuth state saved for MCP server: test-oauth") which was
-        // not caught as UnauthorizedError, causing status to be "failed".
-        expect(serverStatus["test-oauth"]).toBeDefined()
-        expect(serverStatus["test-oauth"].status).toBe("needs_auth")
-      }),
-    ),
-  { config: config("test-oauth") },
+    expect((result.status as Record<string, { status: string }>)["test-oauth"]).toEqual({ status: "needs_auth" })
+  }),
 )
 
-mcpTest.instance("state() generates a new state when none is saved", () =>
+mcpTest.instance("state() generates and persists a new state when none is saved", () =>
   Effect.gen(function* () {
     const auth = yield* McpAuth.Service
     const provider = new McpOAuthProvider(
@@ -204,17 +160,11 @@ mcpTest.instance("state() generates a new state when none is saved", () =>
       auth,
     )
 
-    const entryBefore = yield* McpAuth.use.get("test-state-gen")
-    expect(entryBefore?.oauthState).toBeUndefined()
+    expect((yield* auth.get("test-state-gen"))?.oauthState).toBeUndefined()
 
-    // state() should generate and return a new state, not throw
     const state = yield* Effect.promise(() => provider.state())
-    expect(typeof state).toBe("string")
-    expect(state.length).toBe(64) // 32 bytes as hex
-
-    // The generated state should be persisted
-    const entryAfter = yield* McpAuth.use.get("test-state-gen")
-    expect(entryAfter?.oauthState).toBe(state)
+    expect(state).toHaveLength(64)
+    expect((yield* auth.get("test-state-gen"))?.oauthState).toBe(state)
   }),
 )
 
@@ -229,139 +179,122 @@ mcpTest.instance("state() returns existing state when one is saved", () =>
       auth,
     )
 
-    // Pre-save a state
-    const existingState = "pre-saved-state-value"
-    yield* McpAuth.use.updateOAuthState("test-state-existing", existingState)
-
-    // state() should return the existing state
-    const state = yield* Effect.promise(() => provider.state())
-    expect(state).toBe(existingState)
+    yield* auth.updateOAuthState("test-state-existing", "pre-saved-state-value")
+    expect(yield* Effect.promise(() => provider.state())).toBe("pre-saved-state-value")
   }),
 )
 
-mcpTest.instance(
-  "failed reauthentication preserves existing credentials",
-  () =>
-    Effect.gen(function* () {
-      yield* Effect.addFinalizer(() => Effect.promise(() => McpOAuthCallback.stop()).pipe(Effect.ignore))
-      const mcp = yield* MCP.Service
-      const auth = yield* McpAuth.Service
-      const name = "test-reauth-failure"
-      const url = "https://example.com/mcp"
-      const clientInfo = { clientId: "dynamic-client", clientSecret: "dynamic-secret" }
+mcpTest.instance("pending provider does not expose or overwrite existing credentials before commit", () =>
+  Effect.gen(function* () {
+    const auth = yield* McpAuth.Service
+    const name = "test-pending-credentials"
+    const url = "https://example.com/mcp"
+    const provider = new McpOAuthPendingProvider(name, url, {}, { onRedirect: async () => {} }, auth)
 
-      yield* auth.updateClientInfo(name, clientInfo, url)
-      yield* auth.updateTokens(name, { accessToken: "working-token" }, url)
-      expect((yield* mcp.startAuth(name)).authorizationUrl).toContain("https://auth.example.com/authorize")
-      finishAuthFails = true
+    yield* auth.updateClientInfo(name, { clientId: "old-client" }, url)
+    yield* auth.updateTokens(name, { accessToken: "old-token" }, url)
 
-      expect(yield* mcp.finishAuth(name, "invalid-code")).toEqual({
-        status: "failed",
-        error: "OAuth completion failed: Token exchange failed",
-      })
-      const entry = yield* auth.get(name)
-      expect(entry?.tokens?.accessToken).toBe("working-token")
-      expect(entry?.clientInfo).toEqual(clientInfo)
-    }),
-  { config: config("test-reauth-failure") },
+    expect(yield* Effect.promise(() => provider.clientInformation())).toBeUndefined()
+    expect(yield* Effect.promise(() => provider.tokens())).toBeUndefined()
+    expect((yield* auth.get(name))?.tokens?.accessToken).toBe("old-token")
+    expect((yield* auth.get(name))?.clientInfo?.clientId).toBe("old-client")
+  }),
 )
 
-mcpTest.instance(
-  "successful reauthentication commits replacement credentials",
-  () =>
-    Effect.gen(function* () {
-      yield* Effect.addFinalizer(() => Effect.promise(() => McpOAuthCallback.stop()).pipe(Effect.ignore))
-      const mcp = yield* MCP.Service
-      const auth = yield* McpAuth.Service
-      const name = "test-reauth-success"
-      const url = "https://example.com/mcp"
+mcpTest.instance("failed reauthentication preserves existing credentials", () =>
+  Effect.gen(function* () {
+    yield* stopOAuthCallback
+    const server = yield* serveOAuthMcp()
+    const mcp = yield* MCP.Service
+    const auth = yield* McpAuth.Service
+    const name = "test-reauth-failure"
 
-      yield* auth.updateClientInfo(name, { clientId: "old-client" }, url)
-      yield* auth.updateTokens(name, { accessToken: "old-token" }, url)
-      expect((yield* mcp.startAuth(name)).authorizationUrl).toContain("https://auth.example.com/authorize")
-      expect((yield* auth.get(name))?.tokens?.accessToken).toBe("old-token")
-      finishAuthStoresCredentials = true
-      connectSucceedsImmediately = true
+    yield* auth.updateClientInfo(name, { clientId: "dynamic-client", clientSecret: "dynamic-secret" }, server.url)
+    yield* auth.updateTokens(name, { accessToken: "working-token" }, server.url)
+    yield* mcp.add(name, remote(server.url))
+    expect((yield* mcp.startAuth(name)).authorizationUrl).toContain("/authorize")
 
-      expect((yield* mcp.finishAuth(name, "valid-code")).status).toBe("connected")
-      const entry = yield* auth.get(name)
-      expect(entry?.tokens?.accessToken).toBe("replacement-token")
-      expect(entry?.clientInfo?.clientId).toBe("replacement-client")
-      expect(entry?.serverUrl).toBe(url)
-    }),
-  { config: config("test-reauth-success") },
+    expect(yield* mcp.finishAuth(name, "invalid-code")).toEqual({
+      status: "failed",
+      error: "OAuth completion failed: Token exchange failed",
+    })
+    expect((yield* auth.get(name))?.tokens?.accessToken).toBe("working-token")
+    expect((yield* auth.get(name))?.clientInfo).toMatchObject({
+      clientId: "dynamic-client",
+      clientSecret: "dynamic-secret",
+    })
+  }),
 )
 
-mcpTest.instance(
-  "auth status only reports credentials stored for the configured server URL",
-  () =>
-    Effect.gen(function* () {
-      const mcp = yield* MCP.Service
-      expect(transportCalls).toHaveLength(0)
-      yield* McpAuth.use.updateTokens("test-status-url", { accessToken: "old-token" }, "https://old.example.com/mcp")
+mcpTest.instance("successful reauthentication commits replacement credentials", () =>
+  Effect.gen(function* () {
+    yield* stopOAuthCallback
+    const server = yield* serveOAuthMcp()
+    const mcp = yield* MCP.Service
+    const auth = yield* McpAuth.Service
+    const name = "test-reauth-success"
 
-      expect(yield* mcp.getAuthStatus("test-status-url")).toBe("not_authenticated")
+    yield* auth.updateClientInfo(name, { clientId: "old-client" }, server.url)
+    yield* auth.updateTokens(name, { accessToken: "old-token" }, server.url)
+    yield* mcp.add(name, remote(server.url))
+    expect((yield* mcp.startAuth(name)).authorizationUrl).toContain("/authorize")
+    expect((yield* auth.get(name))?.tokens?.accessToken).toBe("old-token")
 
-      yield* McpAuth.use.updateTokens("test-status-url", { accessToken: "current-token" }, "https://example.com/mcp")
-      expect(yield* mcp.getAuthStatus("test-status-url")).toBe("authenticated")
-
-      yield* McpAuth.use.updateTokens(
-        "test-status-url",
-        { accessToken: "expired-token", expiresAt: 1 },
-        "https://example.com/mcp",
-      )
-      expect(yield* mcp.getAuthStatus("test-status-url")).toBe("expired")
-      expect(transportCalls).toHaveLength(0)
-    }),
-  { config: config("test-status-url") },
+    expect((yield* mcp.finishAuth(name, "valid-code")).status).toBe("connected")
+    const entry = yield* auth.get(name)
+    expect(entry?.tokens?.accessToken).toBe("replacement-token")
+    expect(entry?.clientInfo?.clientId).toBe("replacement-client")
+    expect(entry?.serverUrl).toBe(server.url)
+  }),
 )
 
-mcpTest.instance(
-  "authenticate() stores a connected client when auth completes without redirect",
-  () =>
-    MCP.Service.use((mcp) =>
-      Effect.gen(function* () {
-        const added = yield* mcp.add("test-oauth-connect", {
-          type: "remote",
-          url: "https://example.com/mcp",
-        })
-        const before = added.status as Record<string, { status: string; error?: string }>
-        expect(before["test-oauth-connect"]?.status).toBe("needs_auth")
+mcpTest.instance("auth status only reports credentials stored for the configured server URL", () =>
+  Effect.gen(function* () {
+    const mcp = yield* MCP.Service
+    yield* mcp.add("test-status-url", remote("https://example.com/mcp", false))
+    yield* McpAuth.use.updateTokens("test-status-url", { accessToken: "old-token" }, "https://old.example.com/mcp")
 
-        simulateAuthFlow = false
-        connectSucceedsImmediately = true
+    expect(yield* mcp.getAuthStatus("test-status-url")).toBe("not_authenticated")
 
-        const result = yield* mcp.authenticate("test-oauth-connect")
-        expect(result.status).toBe("connected")
+    yield* McpAuth.use.updateTokens("test-status-url", { accessToken: "current-token" }, "https://example.com/mcp")
+    expect(yield* mcp.getAuthStatus("test-status-url")).toBe("authenticated")
 
-        const after = yield* mcp.status()
-        expect(after["test-oauth-connect"]?.status).toBe("connected")
-      }),
-    ),
-  { config: config("test-oauth-connect") },
+    yield* McpAuth.use.updateTokens(
+      "test-status-url",
+      { accessToken: "expired-token", expiresAt: 1 },
+      "https://example.com/mcp",
+    )
+    expect(yield* mcp.getAuthStatus("test-status-url")).toBe("expired")
+  }),
 )
 
-mcpTest.instance(
-  "authenticate() connects a resource-only server without listing tools",
-  () =>
-    MCP.Service.use((mcp) =>
-      Effect.gen(function* () {
-        const added = yield* mcp.add("test-oauth-resources", {
-          type: "remote",
-          url: "https://example.com/mcp",
-        })
-        const before = added.status as Record<string, { status: string }>
-        expect(before["test-oauth-resources"]?.status).toBe("needs_auth")
+mcpTest.instance("authenticate() stores a connected client when auth completes without redirect", () =>
+  Effect.gen(function* () {
+    yield* stopOAuthCallback
+    const server = yield* serveOAuthMcp()
+    const mcp = yield* MCP.Service
+    const name = "test-oauth-connect"
+    const added = yield* mcp.add(name, remote(server.url))
+    expect((added.status as Record<string, { status: string }>)[name]?.status).toBe("needs_auth")
 
-        simulateAuthFlow = false
-        connectSucceedsImmediately = true
-        serverCapabilities = { resources: {} }
-
-        const result = yield* mcp.authenticate("test-oauth-resources")
-        expect(result.status).toBe("connected")
-        expect(listToolsCalls).toBe(0)
-        expect(Object.keys(yield* mcp.resources())).toEqual(["test-oauth-resources:docs://readme"])
-      }),
-    ),
-  { config: config("test-oauth-resources") },
+    server.allowAnonymous()
+    expect((yield* mcp.authenticate(name)).status).toBe("connected")
+    expect((yield* mcp.status())[name]?.status).toBe("connected")
+  }),
+)
+
+mcpTest.instance("authenticate() connects a resource-only server without listing tools", () =>
+  Effect.gen(function* () {
+    yield* stopOAuthCallback
+    const server = yield* serveOAuthMcp({ capabilities: "resources" })
+    const mcp = yield* MCP.Service
+    const name = "test-oauth-resources"
+    const added = yield* mcp.add(name, remote(server.url))
+    expect((added.status as Record<string, { status: string }>)[name]?.status).toBe("needs_auth")
+
+    server.allowAnonymous()
+    expect((yield* mcp.authenticate(name)).status).toBe("connected")
+    expect(server.listToolsCalls()).toBe(0)
+    expect(Object.keys(yield* mcp.resources())).toEqual([`${name}:docs://readme`])
+  }),
 )

packages/opencode/test/mcp/oauth-browser.test.ts

@@ -1,152 +1,138 @@
-import { expect, mock, beforeEach } from "bun:test"
-import { EventEmitter } from "events"
+import { expect } from "bun:test"
+import { Server } from "@modelcontextprotocol/sdk/server/index.js"
+import { WebStandardStreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/webStandardStreamableHttp.js"
+import { ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
 import { Deferred, Effect, Layer, Option } from "effect"
+import { Config } from "../../src/config/config"
+import { EventV2Bridge } from "../../src/event-v2-bridge"
+import { McpAuth } from "../../src/mcp/auth"
+import { McpBrowser } from "../../src/mcp/browser"
+import { MCP } from "../../src/mcp/index"
+import { McpOAuthCallback } from "../../src/mcp/oauth-callback"
 import { awaitWithTimeout, testEffect } from "../lib/effect"
-import type { MCP as MCPNS } from "../../src/mcp/index"
 
-// Track open() calls and control failure behavior
-let openShouldFail = false
-let openCalledWith: string | undefined
-let openDeferred: Deferred.Deferred<string> | undefined
+const browsers = new Map<string, { opened: Deferred.Deferred<string>; fail: boolean }>()
 
-void mock.module("open", () => ({
-  default: async (url: string) => {
-    openCalledWith = url
-    if (openDeferred) Effect.runSync(Deferred.succeed(openDeferred, url).pipe(Effect.ignore))
-
-    // Return a mock subprocess that emits an error if openShouldFail is true
-    const subprocess = new EventEmitter()
-    if (openShouldFail) {
-      // Emit error asynchronously like a real subprocess would
-      setTimeout(() => {
-        subprocess.emit("error", new Error("spawn xdg-open ENOENT"))
-      }, 10)
-    }
-    return subprocess
-  },
-}))
-
-// Mock UnauthorizedError
-class MockUnauthorizedError extends Error {
-  constructor() {
-    super("Unauthorized")
-    this.name = "UnauthorizedError"
-  }
-}
-
-// Track what options were passed to each transport constructor
-const transportCalls: Array<{
-  type: "streamable" | "sse"
-  url: string
-  options: { authProvider?: unknown; requestInit?: RequestInit }
-}> = []
-
-// Mock the transport constructors
-void mock.module("@modelcontextprotocol/sdk/client/streamableHttp.js", () => ({
-  StreamableHTTPClientTransport: class MockStreamableHTTP {
-    url: string
-    authProvider: { redirectToAuthorization?: (url: URL) => Promise<void> } | undefined
-    constructor(
-      url: URL,
-      options?: { authProvider?: { redirectToAuthorization?: (url: URL) => Promise<void> }; requestInit?: RequestInit },
-    ) {
-      this.url = url.toString()
-      this.authProvider = options?.authProvider
-      transportCalls.push({
-        type: "streamable",
-        url: url.toString(),
-        options: options ?? {},
-      })
-    }
-    async start() {
-      // Simulate OAuth redirect by calling the authProvider's redirectToAuthorization
-      if (this.authProvider?.redirectToAuthorization) {
-        await this.authProvider.redirectToAuthorization(new URL("https://auth.example.com/authorize?client_id=test"))
-      }
-      throw new MockUnauthorizedError()
-    }
-    async finishAuth(_code: string) {
-      // Mock successful auth completion
-    }
-  },
-}))
-
-void mock.module("@modelcontextprotocol/sdk/client/sse.js", () => ({
-  SSEClientTransport: class MockSSE {
-    constructor(url: URL) {
-      transportCalls.push({
-        type: "sse",
-        url: url.toString(),
-        options: {},
-      })
-    }
-    async start() {
-      throw new Error("Mock SSE transport cannot connect")
-    }
-  },
-}))
-
-// Mock the MCP SDK Client to trigger OAuth flow
-void mock.module("@modelcontextprotocol/sdk/client/index.js", () => ({
-  Client: class MockClient {
-    setRequestHandler() {}
-
-    async connect(transport: { start: () => Promise<void> }) {
-      await transport.start()
-    }
-
-    getServerCapabilities() {
-      return { tools: {} }
-    }
-  },
-}))
-
-// Mock UnauthorizedError in the auth module
-void mock.module("@modelcontextprotocol/sdk/client/auth.js", () => ({
-  UnauthorizedError: MockUnauthorizedError,
-}))
-
-beforeEach(() => {
-  openShouldFail = false
-  openCalledWith = undefined
-  openDeferred = undefined
-  transportCalls.length = 0
-})
-
-// Import modules after mocking
-const { MCP } = await import("../../src/mcp/index")
-const { EventV2Bridge } = await import("../../src/event-v2-bridge")
-const { Config } = await import("../../src/config/config")
-const { McpAuth } = await import("../../src/mcp/auth")
-const { McpOAuthCallback } = await import("../../src/mcp/oauth-callback")
-const { FSUtil } = await import("@opencode-ai/core/fs-util")
-const { CrossSpawnSpawner } = await import("@opencode-ai/core/cross-spawn-spawner")
-const mcpTest = testEffect(
-  LayerNode.compile(
-    LayerNode.group([MCP.node, McpAuth.node, EventV2Bridge.node, Config.node, CrossSpawnSpawner.node, FSUtil.node]),
-  ),
+const browserLayer = Layer.succeed(
+  McpBrowser.Service,
+  McpBrowser.Service.of({
+    open: (url) =>
+      Effect.gen(function* () {
+        const browser = browsers.get(new URL(url).origin)
+        if (!browser) return yield* Effect.fail(new Error(`Unexpected browser URL: ${url}`))
+        Deferred.doneUnsafe(browser.opened, Effect.succeed(url))
+        if (browser.fail) return yield* Effect.fail(new Error("spawn xdg-open ENOENT"))
+        yield* Effect.tryPromise({
+          try: () => fetch(url).then((response) => response.body?.cancel()),
+          catch: (error) => (error instanceof Error ? error : new Error(String(error))),
+        })
+      }),
+  }),
 )
-const service = MCP.Service as unknown as Effect.Effect<MCPNS.Interface, never, never>
 
-const config = (name: string, headers?: Record<string, string>) => ({
-  mcp: {
-    [name]: {
-      type: "remote" as const,
-      url: "https://example.com/mcp",
-      headers,
-    },
-  },
-})
+const mcpTest = testEffect(
+  LayerNode.compile(LayerNode.group([MCP.node, McpAuth.node, EventV2Bridge.node, Config.node]), [
+    [McpBrowser.node, browserLayer],
+  ]),
+)
+
+const serveOAuthMcp = Effect.acquireRelease(
+  Effect.promise(async () => {
+    const requests: Array<{ pathname: string; headers: Headers }> = []
+    const protocol = new Server({ name: "oauth-browser", version: "1.0.0" }, { capabilities: { tools: {} } })
+    protocol.setRequestHandler(ListToolsRequestSchema, () => Promise.resolve({ tools: [] }))
+    const transport = new WebStandardStreamableHTTPServerTransport({
+      sessionIdGenerator: () => crypto.randomUUID(),
+      enableJsonResponse: true,
+    })
+    await protocol.connect(transport)
+
+    const http = Bun.serve({
+      hostname: "127.0.0.1",
+      port: 0,
+      async fetch(request) {
+        const url = new URL(request.url)
+        requests.push({ pathname: url.pathname, headers: new Headers(request.headers) })
+
+        if (url.pathname === "/mcp") {
+          if (request.headers.get("authorization") === "Bearer test-access-token") {
+            return transport.handleRequest(request)
+          }
+          return new Response("Unauthorized", {
+            status: 401,
+            headers: {
+              "WWW-Authenticate": `Bearer resource_metadata="${url.origin}/.well-known/oauth-protected-resource/mcp", scope="mcp"`,
+            },
+          })
+        }
+
+        if (url.pathname === "/.well-known/oauth-protected-resource/mcp") {
+          return Response.json({
+            resource: `${url.origin}/mcp`,
+            authorization_servers: [url.origin],
+            scopes_supported: ["mcp"],
+          })
+        }
+
+        if (url.pathname === "/.well-known/oauth-authorization-server") {
+          return Response.json({
+            issuer: url.origin,
+            authorization_endpoint: `${url.origin}/authorize`,
+            token_endpoint: `${url.origin}/token`,
+            registration_endpoint: `${url.origin}/register`,
+            scopes_supported: ["mcp"],
+            response_types_supported: ["code"],
+            grant_types_supported: ["authorization_code"],
+            token_endpoint_auth_methods_supported: ["none"],
+            code_challenge_methods_supported: ["S256"],
+          })
+        }
+
+        if (url.pathname === "/register") {
+          const metadata = await request.json()
+          if (!metadata || typeof metadata !== "object") return new Response("Invalid metadata", { status: 400 })
+          return Response.json({ ...metadata, client_id: "test-client" }, { status: 201 })
+        }
+
+        if (url.pathname === "/authorize") {
+          const redirect = new URL(url.searchParams.get("redirect_uri") ?? "")
+          redirect.searchParams.set("code", "test-code")
+          const state = url.searchParams.get("state")
+          if (state) redirect.searchParams.set("state", state)
+          return Response.redirect(redirect, 302)
+        }
+
+        if (url.pathname === "/token") {
+          return Response.json({ access_token: "test-access-token", token_type: "Bearer", scope: "mcp" })
+        }
+
+        return new Response("Not found", { status: 404 })
+      },
+    })
+
+    return {
+      requests,
+      url: new URL("/mcp", http.url).toString(),
+      close: async () => {
+        await http.stop(true)
+        await protocol.close()
+      },
+    }
+  }),
+  (server) => Effect.promise(server.close),
+)
 
 const withCallbackStop = Effect.addFinalizer(() => Effect.promise(() => McpOAuthCallback.stop()).pipe(Effect.ignore))
 
-const trackBrowserOpen = Effect.gen(function* () {
-  const opened = yield* Deferred.make<string>()
-  openDeferred = opened
-  yield* Effect.addFinalizer(() => Effect.sync(() => (openDeferred = undefined)))
-  return opened
-})
+const trackBrowserOpen = (url: string, fail = false) =>
+  Effect.gen(function* () {
+    const origin = new URL(url).origin
+    const opened = yield* Deferred.make<string>()
+    browsers.set(origin, { opened, fail })
+    yield* Effect.addFinalizer(() => Effect.sync(() => browsers.delete(origin)))
+    return opened
+  })
 
 const trackBrowserOpenFailed = Effect.gen(function* () {
   const events = yield* EventV2Bridge.Service
@@ -160,84 +146,80 @@ const trackBrowserOpenFailed = Effect.gen(function* () {
   return event
 })
 
-const authenticateScoped = (name: string, onAuthorization?: (authorizationUrl: string) => void) =>
+const addServer = Effect.fnUntraced(function* (name: string, url: string, headers?: Record<string, string>) {
+  const mcp = yield* MCP.Service
+  const result = yield* mcp.add(name, { type: "remote", url, headers })
+  expect(result.status).toMatchObject({ [name]: { status: "needs_auth" } })
+  return mcp
+})
+
+mcpTest.instance("BrowserOpenFailed event is published when browser launch fails", () =>
   Effect.gen(function* () {
-    const mcp = yield* service
-    yield* mcp.authenticate(name, onAuthorization).pipe(
-      Effect.ignore,
-      Effect.catchCause(() => Effect.void),
-      Effect.forkScoped,
+    yield* withCallbackStop
+    const server = yield* serveOAuthMcp
+    yield* trackBrowserOpen(server.url, true)
+
+    const event = yield* trackBrowserOpenFailed
+    const mcp = yield* addServer("test-oauth-server", server.url)
+    yield* mcp.authenticate("test-oauth-server").pipe(Effect.ignore, Effect.forkScoped)
+
+    const failure = yield* awaitWithTimeout(
+      Deferred.await(event),
+      "Timed out waiting for BrowserOpenFailed event",
+      "5 seconds",
     )
-  })
 
-mcpTest.instance(
-  "BrowserOpenFailed event is published when open() throws",
-  () =>
-    Effect.gen(function* () {
-      yield* withCallbackStop
-      openShouldFail = true
-
-      const event = yield* trackBrowserOpenFailed
-      yield* authenticateScoped("test-oauth-server")
-
-      const failure = yield* awaitWithTimeout(
-        Deferred.await(event),
-        "Timed out waiting for BrowserOpenFailed event",
-        "5 seconds",
-      )
-
-      expect(failure.mcpName).toBe("test-oauth-server")
-      expect(failure.url).toContain("https://")
-    }),
-  { config: config("test-oauth-server") },
+    expect(failure.mcpName).toBe("test-oauth-server")
+    expect(failure.url).toStartWith(new URL("/authorize", server.url).toString())
+  }),
 )
 
-mcpTest.instance(
-  "BrowserOpenFailed event is NOT published when open() succeeds",
-  () =>
-    Effect.gen(function* () {
-      yield* withCallbackStop
-      openShouldFail = false
+mcpTest.instance("BrowserOpenFailed event is not published when browser launch succeeds", () =>
+  Effect.gen(function* () {
+    yield* withCallbackStop
+    const server = yield* serveOAuthMcp
 
-      const opened = yield* trackBrowserOpen
-      const event = yield* trackBrowserOpenFailed
-      yield* authenticateScoped("test-oauth-server-2")
+    const opened = yield* trackBrowserOpen(server.url)
+    const event = yield* trackBrowserOpenFailed
+    const mcp = yield* addServer("test-oauth-server-2", server.url)
+    const status = yield* awaitWithTimeout(
+      mcp.authenticate("test-oauth-server-2"),
+      "Timed out completing OAuth authentication",
+      "5 seconds",
+    )
+    const url = yield* Deferred.await(opened)
+    const failure = yield* Deferred.await(event).pipe(Effect.timeoutOption("700 millis"))
 
-      yield* awaitWithTimeout(Deferred.await(opened), "Timed out waiting for open()", "5 seconds")
-      const failure = yield* Deferred.await(event).pipe(Effect.timeoutOption("700 millis"))
-
-      expect(failure).toEqual(Option.none())
-      expect(openCalledWith).toBeDefined()
-    }),
-  { config: config("test-oauth-server-2") },
+    expect(status).toEqual({ status: "connected" })
+    expect(failure).toEqual(Option.none())
+    expect(new URL(url).origin).toBe(new URL(server.url).origin)
+  }),
 )
 
-mcpTest.instance(
-  "open() is called with the authorization URL",
-  () =>
-    Effect.gen(function* () {
-      yield* withCallbackStop
-      openShouldFail = false
-      openCalledWith = undefined
+mcpTest.instance("browser launch receives the discovered authorization URL", () =>
+  Effect.gen(function* () {
+    yield* withCallbackStop
+    const server = yield* serveOAuthMcp
 
-      const opened = yield* trackBrowserOpen
-      const event = yield* trackBrowserOpenFailed
-      const authorization = yield* Deferred.make<string>()
-      yield* authenticateScoped("test-oauth-server-3", (url) => Deferred.doneUnsafe(authorization, Effect.succeed(url)))
+    const opened = yield* trackBrowserOpen(server.url)
+    const authorization = yield* Deferred.make<string>()
+    const mcp = yield* addServer("test-oauth-server-3", server.url, { "X-Custom-Header": "custom-value" })
+    const status = yield* awaitWithTimeout(
+      mcp.authenticate("test-oauth-server-3", (url) => Deferred.doneUnsafe(authorization, Effect.succeed(url))),
+      "Timed out completing OAuth authentication",
+      "5 seconds",
+    )
+    const url = yield* Deferred.await(opened)
+    const authorizationUrl = yield* Deferred.await(authorization)
 
-      const url = yield* awaitWithTimeout(Deferred.await(opened), "Timed out waiting for open()", "5 seconds")
-      const authorizationUrl = yield* awaitWithTimeout(
-        Deferred.await(authorization),
-        "Timed out waiting for authorization URL",
-        "5 seconds",
-      )
-      const failure = yield* Deferred.await(event).pipe(Effect.timeoutOption("700 millis"))
-
-      expect(failure).toEqual(Option.none())
-      expect(authorizationUrl).toBe(url)
-      expect(typeof url).toBe("string")
-      expect(url).toContain("https://")
-      expect(transportCalls.at(-1)?.options.requestInit?.headers).toEqual({ "X-Custom-Header": "custom-value" })
-    }),
-  { config: config("test-oauth-server-3", { "X-Custom-Header": "custom-value" }) },
+    expect(status).toEqual({ status: "connected" })
+    expect(authorizationUrl).toBe(url)
+    expect(new URL(url).pathname).toBe("/authorize")
+    expect(new URL(url).searchParams.get("client_id")).toBe("test-client")
+    expect(
+      server.requests.some(
+        (request) => request.pathname === "/mcp" && request.headers.get("x-custom-header") === "custom-value",
+      ),
+    ).toBe(true)
+  }),
 )

packages/opencode/test/plugin/github-copilot-models.test.ts

@@ -187,6 +187,44 @@ test("converts Copilot AIC token prices to USD per million tokens", async () =>
   expect(models["ignored-non-chat-record"]).toBeUndefined()
 })
 
+test("records Copilot advertised responses endpoint for non-GPT model IDs", async () => {
+  globalThis.fetch = mock(() =>
+    Promise.resolve(
+      new Response(
+        JSON.stringify({
+          data: [
+            {
+              model_picker_enabled: true,
+              id: "mai-code-1-flash-picker",
+              name: "MAI-Code-1-Flash",
+              version: "mai-code-1-flash-picker",
+              supported_endpoints: ["/responses"],
+              capabilities: {
+                family: "oswe-vscode-modelD",
+                limits: {
+                  max_context_window_tokens: 256000,
+                  max_output_tokens: 128000,
+                  max_prompt_tokens: 128000,
+                },
+                supports: {
+                  streaming: true,
+                  structured_outputs: true,
+                  tool_calls: true,
+                },
+              },
+            },
+          ],
+        }),
+        { status: 200 },
+      ),
+    ),
+  ) as unknown as typeof fetch
+
+  const model = (await CopilotModels.get("https://api.githubcopilot.com")).models["mai-code-1-flash-picker"]
+
+  expect("endpoint" in model.api ? model.api.endpoint : undefined).toBe("responses")
+})
+
 test("clears existing variants so refreshed models calculate provider-specific variants", async () => {
   globalThis.fetch = mock(() =>
     Promise.resolve(

packages/opencode/test/provider/transform.test.ts

@@ -4638,12 +4638,12 @@ describe("ProviderTransform.smallOptions - gpt-5 chat/search", () => {
   }
 })
 
-test("ProviderTransform.smallOptions disables OpenRouter reasoning when the weakest effort is low", () => {
+test("ProviderTransform.smallOptions preserves the weakest OpenRouter reasoning effort", () => {
   expect(
     ProviderTransform.smallOptions({
       providerID: "openrouter",
       api: {
-        id: "anthropic/claude-sonnet-4.6",
+        id: "google/gemini-3.5-flash",
         npm: "@openrouter/ai-sdk-provider",
       },
       variants: {
@@ -4652,7 +4652,7 @@ test("ProviderTransform.smallOptions disables OpenRouter reasoning when the weak
         high: { reasoning: { effort: "high" } },
       },
     } as any),
-  ).toEqual({ reasoning: { effort: "none" } })
+  ).toEqual({ reasoning: { effort: "low" } })
 })
 
 describe("ProviderTransform.smallOptions - google thinking controls", () => {

packages/opencode/test/server/httpapi-exercise/index.ts

@@ -1630,33 +1630,21 @@ const scenarios: Scenario[] = [
         const session = yield* ctx.session({ title: "Summarize session" })
         yield* ctx.message(session.id, { text: "summarize this work" })
         const summary = [
-          "## Goal",
+          "## Objective",
           "- Exercise session summarize.",
           "",
-          "## Constraints & Preferences",
+          "## Important Details",
           "- Use fake LLM.",
-          "",
-          "## Progress",
-          "### Done",
-          "- Summary generated.",
-          "",
-          "### In Progress",
-          "- (none)",
-          "",
-          "### Blocked",
-          "- (none)",
-          "",
-          "## Key Decisions",
           "- Keep route local.",
+          "- Test fixture: test/server/httpapi-exercise/index.ts.",
           "",
-          "## Next Steps",
-          "- (none)",
+          "## Work State",
+          "- Completed: Summary generated.",
+          "- Active: (none)",
+          "- Blocked: (none)",
           "",
-          "## Critical Context",
-          "- Test fixture.",
-          "",
-          "## Relevant Files",
-          "- test/server/httpapi-exercise/index.ts: scenario",
+          "## Next Move",
+          "1. (none)",
         ].join("\n")
         yield* ctx.llmText(summary)
         yield* ctx.llmText(summary)

packages/opencode/test/server/httpapi-session.test.ts

@@ -21,6 +21,7 @@ import { InstanceStore } from "../../src/project/instance-store"
 import { Project } from "../../src/project/project"
 import { HttpApiApp } from "../../src/server/routes/instance/httpapi/server"
 import * as HttpSessionError from "../../src/server/routes/instance/httpapi/handlers/session-errors"
+import { ExperimentalPaths } from "../../src/server/routes/instance/httpapi/groups/experimental"
 import { SessionPaths } from "../../src/server/routes/instance/httpapi/groups/session"
 import { Session } from "@/session/session"
 import { MessageID, PartID, SessionID, type SessionID as SessionIDType } from "../../src/session/schema"
@@ -879,6 +880,78 @@ describe("session HttpApi", () => {
     { git: true, config: { formatter: false, lsp: false } },
   )
 
+  it.instance(
+    "lists sessions created through an equivalent directory hint",
+    () =>
+      Effect.gen(function* () {
+        const test = yield* TestInstance
+        const hint = test.directory + path.sep
+        const headers = { "x-opencode-directory": hint, "content-type": "application/json" }
+        const created = yield* requestJson<Session.Info>(SessionPaths.create, {
+          method: "POST",
+          headers,
+          body: JSON.stringify({ title: "hinted" }),
+        })
+
+        const query = new URLSearchParams({ directory: hint, roots: "true" })
+        const listed = yield* requestJson<Session.Info[]>(`${SessionPaths.list}?${query}`, { headers })
+        expect(listed.map((item) => item.id)).toContain(created.id)
+
+        const globalQuery = new URLSearchParams({ directory: hint })
+        const global = yield* requestJson<Session.Info[]>(`${ExperimentalPaths.session}?${globalQuery}`, { headers })
+        expect(global.map((item) => item.id)).toContain(created.id)
+      }),
+    { git: true, config: { formatter: false, lsp: false, share: "disabled" } },
+  )
+
+  it.instance(
+    "lists Windows sessions for equivalent directory spellings",
+    () =>
+      Effect.gen(function* () {
+        if (process.platform !== "win32") return
+        const test = yield* TestInstance
+        const headers = { "x-opencode-directory": test.directory, "content-type": "application/json" }
+        const created = yield* requestJson<Session.Info>(SessionPaths.create, {
+          method: "POST",
+          headers,
+          body: JSON.stringify({ title: "windows spelling" }),
+        })
+
+        const forwardSlashes = test.directory.replaceAll("\\", "/")
+        const lowercaseDrive = test.directory.replace(/^[A-Z]:/, (drive) => drive.toLowerCase())
+        const trailingSeparator = `${test.directory}\\`
+        for (const spelling of [forwardSlashes, lowercaseDrive, trailingSeparator]) {
+          const query = new URLSearchParams({ directory: spelling, roots: "true" })
+          const listed = yield* requestJson<Session.Info[]>(`${SessionPaths.list}?${query}`, { headers })
+          expect({ spelling, ids: listed.map((item) => item.id) }).toEqual({ spelling, ids: [created.id] })
+        }
+      }),
+    { git: true, config: { formatter: false, lsp: false, share: "disabled" } },
+    { timeout: 15000 },
+  )
+
+  it.instance(
+    "lists Windows sessions created through the global worktree sentinel",
+    () =>
+      Effect.gen(function* () {
+        if (process.platform !== "win32") return
+        const globalWorktreeSentinel = "/"
+        const headers = { "x-opencode-directory": globalWorktreeSentinel, "content-type": "application/json" }
+        const driveRootSession = yield* requestJson<Session.Info>(SessionPaths.create, {
+          method: "POST",
+          headers,
+          body: JSON.stringify({ title: "created at drive root" }),
+        })
+        expect(driveRootSession.directory).toMatch(/^[A-Za-z]:\\$/)
+
+        const query = new URLSearchParams({ directory: globalWorktreeSentinel, roots: "true" })
+        const listed = yield* requestJson<Session.Info[]>(`${SessionPaths.list}?${query}`, { headers })
+        expect(listed.map((item) => item.id)).toContain(driveRootSession.id)
+      }),
+    { git: true, config: { formatter: false, lsp: false, share: "disabled" } },
+    { timeout: 15000 },
+  )
+
   it.instance(
     "serves paginated message link headers",
     () =>

packages/opencode/test/session/compaction.test.ts

@@ -1430,8 +1430,8 @@ describe("session.compaction.process", () => {
         expect(captured).toContain("<previous-summary>")
         expect(captured).toContain("summary one")
         expect(captured.match(/summary one/g)?.length).toBe(1)
-        expect(captured).toContain("## Constraints & Preferences")
-        expect(captured).toContain("## Progress")
+        expect(captured).toContain("## Important Details")
+        expect(captured).toContain("## Work State")
       }).pipe(withCompaction({ llm: stub.llmLayer }))
     },
     { git: true },

packages/opencode/test/session/llm.test.ts

@@ -832,6 +832,80 @@ describe("session.llm.stream", () => {
     },
   )
 
+  const cerebrasFixture = { providerID: "cerebras", modelID: "gpt-oss-120b" }
+  it.instance(
+    "replays Cerebras assistant reasoning using the provider-supported field",
+    () =>
+      Effect.gen(function* () {
+        const fixture = loadFixture(cerebrasFixture.providerID, cerebrasFixture.modelID)
+        const request = waitRequest(
+          "/chat/completions",
+          new Response(createChatStream("Hello"), {
+            status: 200,
+            headers: { "Content-Type": "text/event-stream" },
+          }),
+        )
+
+        const resolved = yield* Provider.use.getModel(
+          ProviderV2.ID.make(cerebrasFixture.providerID),
+          ModelV2.ID.make(fixture.model.id),
+        )
+        const sessionID = SessionID.make("session-test-cerebras-reasoning")
+        const agent = {
+          name: "test",
+          mode: "primary",
+          options: {},
+          permission: [{ permission: "*", pattern: "*", action: "allow" }],
+        } satisfies Agent.Info
+
+        const user = {
+          id: MessageID.make("msg_user-cerebras-reasoning"),
+          sessionID,
+          role: "user",
+          time: { created: Date.now() },
+          agent: agent.name,
+          model: { providerID: ProviderV2.ID.make(cerebrasFixture.providerID), modelID: resolved.id },
+        } satisfies SessionV1.User
+
+        yield* drain({
+          user,
+          sessionID,
+          model: resolved,
+          agent,
+          system: ["You are a helpful assistant."],
+          messages: [
+            { role: "user", content: "Hello" },
+            {
+              role: "assistant",
+              content: [
+                { type: "reasoning", text: "thinking" },
+                { type: "text", text: "Previous answer" },
+              ],
+            },
+            { role: "user", content: "Continue" },
+          ] satisfies ModelMessage[],
+          tools: {},
+        })
+
+        const capture = yield* Effect.promise(() => request)
+        const messages = capture.body.messages as Array<Record<string, unknown>>
+        const assistant = messages.find((msg) => msg.role === "assistant")
+
+        expect(assistant?.reasoning).toBe("thinking")
+        expect(assistant && "reasoning_content" in assistant).toBe(false)
+      }),
+    {
+      config: () => ({
+        enabled_providers: [cerebrasFixture.providerID],
+        provider: {
+          [cerebrasFixture.providerID]: {
+            options: { apiKey: "test-key", baseURL: `${state.server!.url.origin}/v1` },
+          },
+        },
+      }),
+    },
+  )
+
   const alibabaQwenFixture = { providerID: "alibaba", modelID: "qwen-plus" }
   it.instance(
     "service stream cancellation cancels provider response body promptly",

packages/opencode/test/tool/code-mode-integration.test.ts

@@ -0,0 +1,331 @@
+import { beforeAll, describe, expect, test } from "bun:test"
+import { CodeModeTool, describeCatalog } from "@/tool/code-mode"
+import { McpCatalog } from "@/mcp/catalog"
+import { Agent } from "@/agent/agent"
+import { MCP } from "@/mcp"
+import { Plugin } from "@/plugin"
+import { Session } from "@/session/session"
+import { Tool } from "@/tool/tool"
+import * as Truncate from "@/tool/truncate"
+import { MessageID, SessionID } from "@/session/schema"
+import { Server } from "@modelcontextprotocol/sdk/server/index.js"
+import { InMemoryTransport } from "@modelcontextprotocol/sdk/inMemory.js"
+import type { Client } from "@modelcontextprotocol/sdk/client/index.js"
+import {
+  CallToolRequestSchema,
+  LATEST_PROTOCOL_VERSION,
+  ListToolsRequestSchema,
+  type Tool as MCPToolDef,
+} from "@modelcontextprotocol/sdk/types.js"
+import { Cause, Effect, Exit, Layer } from "effect"
+
+const PNG = "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAQAAAC1HAwCAAAAC0lEQVR42mNkYPhfDwAChwGA60e6kgAAAABJRU5ErkJggg=="
+
+const SERVER = "fixtures"
+
+const ctx: Tool.Context = {
+  sessionID: SessionID.make("ses_code-mode-int"),
+  messageID: MessageID.make("msg_code-mode-int"),
+  agent: "build",
+  abort: new AbortController().signal,
+  callID: "call_code_mode_int",
+  messages: [],
+  metadata: () => Effect.void,
+  ask: () => Effect.void,
+}
+
+// Avoid the SDK Client here; other MCP tests mock it process-globally.
+class RawJsonRpcClient {
+  private nextId = 1
+  private pending = new Map<number, { resolve: (value: any) => void; reject: (error: Error) => void }>()
+
+  constructor(private transport: InMemoryTransport) {}
+
+  async connect() {
+    this.transport.onmessage = (message) => {
+      const msg = message as { id?: number; result?: unknown; error?: { message: string } }
+      if (msg.id === undefined) return
+      const entry = this.pending.get(msg.id)
+      if (!entry) return
+      this.pending.delete(msg.id)
+      if (msg.error) entry.reject(new Error(msg.error.message))
+      else entry.resolve(msg.result)
+    }
+    await this.transport.start()
+    await this.request("initialize", {
+      protocolVersion: LATEST_PROTOCOL_VERSION,
+      capabilities: {},
+      clientInfo: { name: "test-client", version: "1.0.0" },
+    })
+    await this.transport.send({ jsonrpc: "2.0", method: "notifications/initialized" })
+  }
+
+  private request(method: string, params: unknown): Promise<any> {
+    const id = this.nextId++
+    const result = new Promise((resolve, reject) => this.pending.set(id, { resolve, reject }))
+    void this.transport.send({ jsonrpc: "2.0", id, method, params } as never)
+    return result
+  }
+
+  listTools() {
+    return this.request("tools/list", {})
+  }
+
+  callTool(params: { name: string; arguments?: Record<string, unknown> }, _schema?: unknown, _options?: unknown) {
+    return this.request("tools/call", params)
+  }
+}
+
+const TOOL_DEFS: MCPToolDef[] = [
+  {
+    name: "get_text",
+    description: "Greet someone and return the greeting as text",
+    inputSchema: { type: "object", properties: { name: { type: "string" } }, required: ["name"] },
+  },
+  {
+    name: "add",
+    description: "Add two numbers and return the structured sum",
+    inputSchema: { type: "object", properties: { a: { type: "number" }, b: { type: "number" } }, required: ["a", "b"] },
+    outputSchema: { type: "object", properties: { sum: { type: "number" } }, required: ["sum"] },
+  },
+  {
+    name: "screenshot",
+    description: "Capture a screenshot and return it as an image",
+    inputSchema: { type: "object", properties: {} },
+  },
+  {
+    name: "boom",
+    description: "A tool that always fails",
+    inputSchema: { type: "object", properties: {} },
+  },
+] as MCPToolDef[]
+
+function handleCall(name: string, args: Record<string, unknown>) {
+  switch (name) {
+    case "get_text":
+      return { content: [{ type: "text", text: `hello ${args.name}` }] }
+    case "add": {
+      const sum = (args.a as number) + (args.b as number)
+      return { content: [{ type: "text", text: String(sum) }], structuredContent: { sum } }
+    }
+    case "screenshot":
+      return { content: [{ type: "image", data: PNG, mimeType: "image/png" }] }
+    case "boom":
+      return { content: [{ type: "text", text: "kaboom" }], isError: true }
+    default:
+      return { content: [{ type: "text", text: `unknown tool ${name}` }], isError: true }
+  }
+}
+
+let tool: Awaited<ReturnType<typeof buildTool>>["tool"]
+let description: string
+
+async function buildTool() {
+  const server = new Server({ name: SERVER, version: "1.0.0" }, { capabilities: { tools: {} } })
+  server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools: TOOL_DEFS }))
+  server.setRequestHandler(CallToolRequestSchema, async (req) =>
+    handleCall(req.params.name, (req.params.arguments ?? {}) as Record<string, unknown>),
+  )
+
+  const [clientTransport, serverTransport] = InMemoryTransport.createLinkedPair()
+  await server.connect(serverTransport)
+  const client = new RawJsonRpcClient(clientTransport)
+  await client.connect()
+
+  const listed = (await client.listTools()).tools as MCPToolDef[]
+  const mcpTools: Record<string, MCP.McpTool> = {}
+  for (const def of listed) {
+    mcpTools[McpCatalog.toolName(SERVER, def.name)] = { def, client: client as unknown as Client }
+  }
+
+  const layer = Layer.mergeAll(
+    Layer.mock(Plugin.Service, {
+      trigger: ((_name: unknown, _input: unknown, output: unknown) =>
+        Effect.succeed(output)) as Plugin.Interface["trigger"],
+    }),
+    Layer.mock(Truncate.Service, {
+      output: (text: string) => Effect.succeed({ content: text, truncated: false as const }),
+    }),
+    Layer.mock(Agent.Service, { get: () => Effect.succeed({ name: "build", permission: [] } as any) }),
+    Layer.mock(Session.Service, { get: () => Effect.succeed({ permission: [] } as any) }),
+    Layer.mock(MCP.Service, {
+      tools: () => Effect.succeed(mcpTools),
+      clients: () => Effect.succeed({ [SERVER]: {} as any }),
+    }),
+  )
+  return {
+    tool: await Effect.runPromise(CodeModeTool.pipe(Effect.flatMap(Tool.init), Effect.provide(layer))),
+    description: describeCatalog(mcpTools, [SERVER]),
+  }
+}
+
+const run = (code: string) => Effect.runPromise(tool.execute({ code }, ctx))
+// Program failures die at the tool boundary; recover the defect for message assertions.
+const runFailed = async (code: string) => {
+  const exit = await Effect.runPromise(tool.execute({ code }, ctx).pipe(Effect.exit))
+  if (Exit.isSuccess(exit)) throw new Error("expected the tool to fail")
+  return Cause.squash(exit.cause) as Error
+}
+
+beforeAll(async () => {
+  const built = await buildTool()
+  tool = built.tool
+  description = built.description
+})
+
+describe("code mode integration (real MCP server)", () => {
+  test("the appended catalog inlines full signatures with real MCP schemas", () => {
+    expect(description).toContain("Available tools (COMPLETE list")
+    expect(description).toContain("- fixtures (4 tools)")
+    expect(description).toContain(
+      "tools.fixtures.add(input: {\n  a: number,\n  b: number,\n}): Promise<{\n  sum: number,\n}>",
+    )
+    expect(description).toContain("tools.fixtures.get_text(input: {\n  name: string,\n}): Promise<unknown>")
+    expect(description).toContain("// Add two numbers and return the structured sum")
+    expect(description).not.toContain("$codemode")
+    expect(description).toContain("## Workflow")
+    expect(description).toContain("Do not infer or normalize tool names")
+    expect(description).toContain("bracket notation and quotes are part of the path")
+    expect(description).not.toContain("total_count")
+  })
+
+  test("calls a text tool and receives its text as the native result", async () => {
+    const out = await run("const r = await tools.fixtures.get_text({ name: 'world' }); return r")
+    expect(out.output).toBe("hello world")
+    expect(out.metadata.toolCalls).toEqual([
+      { tool: "fixtures.get_text", status: "completed", input: { name: "world" } },
+    ])
+    expect(out.attachments).toBeUndefined()
+  })
+
+  test("exposes structured data natively from a tool with an outputSchema", async () => {
+    const out = await run("const r = await tools.fixtures.add({ a: 2, b: 3 }); return r.sum")
+    expect(out.output).toBe("5")
+  })
+
+  test("composes multiple structured calls and returns a plain object", async () => {
+    const out = await run(`
+      const first = await tools.fixtures.add({ a: 1, b: 2 })
+      const second = await tools.fixtures.add({ a: first.sum, b: 10 })
+      return { total: second.sum }
+    `)
+    expect(JSON.parse(out.output)).toEqual({ total: 13 })
+    expect(out.metadata.toolCalls).toEqual([
+      { tool: "fixtures.add", status: "completed", input: { a: 1, b: 2 } },
+      { tool: "fixtures.add", status: "completed", input: { a: 3, b: 10 } },
+    ])
+  })
+
+  test("an image result becomes an execute attachment and a marker in the sandbox", async () => {
+    const out = await run("return await tools.fixtures.screenshot({})")
+    expect(out.output).toBe("[1 image attached to the result]")
+    expect(out.attachments).toEqual([{ type: "file", mime: "image/png", url: `data:image/png;base64,${PNG}` }])
+  })
+
+  test("image bytes never enter the sandbox or the model-facing output", async () => {
+    const out = await run(`
+      const shot = await tools.fixtures.screenshot({})
+      return { sawMarker: typeof shot === 'string' && shot.includes('attached'), value: shot }
+    `)
+    expect(JSON.parse(out.output)).toEqual({
+      sawMarker: true,
+      value: "[1 image attached to the result]",
+    })
+    expect(out.output).not.toContain(PNG)
+    expect(out.attachments).toHaveLength(1)
+  })
+
+  test("attachments accumulate even when the program returns something else", async () => {
+    const out = await run("await tools.fixtures.screenshot({}); return 'captured'")
+    expect(out.output).toBe("captured")
+    expect(out.attachments).toHaveLength(1)
+  })
+
+  test("runs calls in parallel and accumulates every attachment", async () => {
+    const out = await run(`
+      const both = await Promise.all([tools.fixtures.screenshot({}), tools.fixtures.screenshot({})])
+      return 'two shots: ' + both.length
+    `)
+    expect(out.output).toBe("two shots: 2")
+    expect(out.attachments).toHaveLength(2)
+    expect(out.metadata.toolCalls.map((c) => c.tool)).toEqual(["fixtures.screenshot", "fixtures.screenshot"])
+  })
+
+  test("propagates an MCP isError into the program as a catchable error", async () => {
+    const out = await run("try { await tools.fixtures.boom({}) } catch (e) { return 'caught: ' + e.message }")
+    expect(out.output).toBe("caught: kaboom")
+  })
+
+  test("an uncaught MCP error surfaces as a failed execution", async () => {
+    const error = await runFailed("await tools.fixtures.boom({}); return 'unreachable'")
+    expect(error.message).toContain("kaboom")
+  })
+
+  test("console output is captured and appended as a Logs section after the result", async () => {
+    const out = await run(`
+      console.log("looking up", { name: "world" })
+      const r = await tools.fixtures.get_text({ name: "world" })
+      console.warn("got", r)
+      return r
+    `)
+    expect(out.output).toBe('hello world\n\nLogs:\nlooking up {"name":"world"}\n[warn] got hello world')
+    expect(out.metadata.error).toBeUndefined()
+  })
+
+  test("console output is preserved on the error path", async () => {
+    const error = await runFailed(`
+      console.log("before the throw")
+      await tools.fixtures.boom({})
+      return "unreachable"
+    `)
+    expect(error.message).toContain("kaboom")
+    expect(error.message).toContain("Logs:\nbefore the throw")
+  })
+
+  test("a program that logs nothing gets no Logs section", async () => {
+    const out = await run("return 'quiet'")
+    expect(out.output).toBe("quiet")
+    expect(out.output).not.toContain("Logs:")
+  })
+
+  test("console does not consume the tool-call metadata (logging is not a tool call)", async () => {
+    const out = await run("console.log('hi'); console.error('bye'); return 'ok'")
+    expect(out.output).toBe("ok\n\nLogs:\nhi\n[error] bye")
+    expect(out.metadata.toolCalls).toEqual([])
+  })
+
+  test("asks permission for each MCP call, keyed by the flat catalog name", async () => {
+    const asked: string[] = []
+    const permCtx: Tool.Context = { ...ctx, ask: (req: any) => Effect.sync(() => void asked.push(req.permission)) }
+    await Effect.runPromise(
+      tool.execute(
+        {
+          code: `
+            await tools.fixtures.add({ a: 1, b: 1 })
+            await tools.fixtures.get_text({ name: 'x' })
+            return 'done'
+          `,
+        },
+        permCtx,
+      ),
+    )
+    expect(asked).toEqual(["fixtures_add", "fixtures_get_text"])
+  })
+
+  test("streams running/completed metadata for child calls over a real transport", async () => {
+    const snapshots: Array<{ toolCalls: { tool: string; status: string; input?: Record<string, unknown> }[] }> = []
+    const recordingCtx: Tool.Context = {
+      ...ctx,
+      metadata: (val: any) => Effect.sync(() => void snapshots.push(val.metadata)),
+    }
+    await Effect.runPromise(
+      tool.execute({ code: "await tools.fixtures.add({ a: 1, b: 2 }); return 'done'" }, recordingCtx),
+    )
+    expect(snapshots).toContainEqual({
+      toolCalls: [{ tool: "fixtures.add", status: "running", input: { a: 1, b: 2 } }],
+    })
+    expect(snapshots).toContainEqual({
+      toolCalls: [{ tool: "fixtures.add", status: "completed", input: { a: 1, b: 2 } }],
+    })
+  })
+})

packages/opencode/test/tool/code-mode.test.ts

@@ -0,0 +1,730 @@
+import { describe, expect, test } from "bun:test"
+import { CODE_MODE_TOOL, CodeModeTool, Parameters, describeCatalog } from "@/tool/code-mode"
+import type { Tool as MCPToolDef } from "@modelcontextprotocol/sdk/types.js"
+import type { PermissionV1 } from "@opencode-ai/core/v1/permission"
+import { Agent } from "@/agent/agent"
+import { MCP } from "@/mcp"
+import { Permission } from "@/permission"
+import { Plugin } from "@/plugin"
+import { Session } from "@/session/session"
+import { Tool } from "@/tool/tool"
+import * as Truncate from "@/tool/truncate"
+import { MessageID, SessionID } from "@/session/schema"
+import { Cause, Effect, Exit, Layer, Schema } from "effect"
+
+const ctx: Tool.Context = {
+  sessionID: SessionID.make("ses_code-mode"),
+  messageID: MessageID.make("msg_code-mode"),
+  agent: "build",
+  abort: new AbortController().signal,
+  callID: "call_code_mode",
+  messages: [],
+  metadata: () => Effect.void,
+  ask: () => Effect.void,
+}
+
+function mcpTool(
+  name: string,
+  handler: (args: Record<string, unknown>) => unknown,
+  inputSchema: Record<string, unknown> = { type: "object", properties: {} },
+  outputSchema?: Record<string, unknown>,
+): MCP.McpTool {
+  return {
+    def: { name, description: name, inputSchema, ...(outputSchema ? { outputSchema } : {}) } as MCPToolDef,
+    client: {
+      callTool: async (params: { arguments?: Record<string, unknown> }) => handler(params.arguments ?? {}),
+    } as unknown as MCP.McpTool["client"],
+  }
+}
+
+function harness(input: {
+  mcpTools: Record<string, MCP.McpTool>
+  servers: string[]
+  permission?: PermissionV1.Rule[]
+  trigger?: Plugin.Interface["trigger"]
+}) {
+  return Layer.mergeAll(
+    Layer.mock(Plugin.Service, {
+      trigger: input.trigger ?? (((_name, _input, output) => Effect.succeed(output)) as Plugin.Interface["trigger"]),
+    }),
+    Layer.mock(Truncate.Service, {
+      output: (text: string) => Effect.succeed({ content: text, truncated: false as const }),
+    }),
+    Layer.mock(Agent.Service, {
+      get: () => Effect.succeed({ name: "build", permission: input.permission ?? [] } as any),
+    }),
+    Layer.mock(Session.Service, {
+      get: () => Effect.succeed({ permission: [] } as any),
+    }),
+    Layer.mock(MCP.Service, {
+      tools: () => Effect.succeed(input.mcpTools),
+      clients: () => Effect.succeed(Object.fromEntries(input.servers.map((name) => [name, {} as any]))),
+    }),
+  )
+}
+
+function serverNames(mcpTools: Record<string, MCP.McpTool>, servers?: string[]) {
+  return servers ?? [...new Set(Object.keys(mcpTools).map((key) => key.split("_")[0]!))]
+}
+
+function build(
+  mcpTools: Record<string, MCP.McpTool>,
+  servers?: string[],
+  permission?: PermissionV1.Rule[],
+  trigger?: Plugin.Interface["trigger"],
+) {
+  const names = serverNames(mcpTools, servers)
+  return Effect.runPromise(
+    CodeModeTool.pipe(
+      Effect.flatMap(Tool.init),
+      Effect.provide(harness({ mcpTools, servers: names, permission, trigger })),
+    ),
+  )
+}
+
+function describeFor(mcpTools: Record<string, MCP.McpTool>, servers?: string[], permission: PermissionV1.Rule[] = []) {
+  return describeCatalog(Permission.visibleTools(mcpTools, permission), serverNames(mcpTools, servers))
+}
+
+// Program failures die at the tool boundary; recover the defect for message assertions.
+async function failure(effect: Effect.Effect<unknown>) {
+  const exit = await Effect.runPromise(effect.pipe(Effect.exit))
+  if (Exit.isSuccess(exit)) throw new Error("expected the tool to fail")
+  return Cause.squash(exit.cause) as Error
+}
+
+describe("code mode execute", () => {
+  test("defines execute input with an Effect schema", async () => {
+    const decode = Schema.decodeUnknownEffect(Parameters)
+    await expect(Effect.runPromise(decode({ code: "return 1" }))).resolves.toEqual({ code: "return 1" })
+    await expect(Effect.runPromise(decode({}))).rejects.toThrow()
+    expect(Schema.toJsonSchemaDocument(Parameters).schema).toMatchObject({
+      properties: {
+        code: {
+          description: "Script body executed by the confined interpreter.",
+        },
+      },
+    })
+  })
+
+  test("groups multi-underscore server names by longest matching prefix", () => {
+    const description = describeFor({ my_server_do_thing: mcpTool("do_thing", () => "") }, ["my_server"])
+    expect(description).toContain("- my_server (1 tool)")
+    expect(description).toContain("tools.my_server.do_thing(")
+  })
+
+  test("groupByServer uses the whole key as the server name when it has no underscore", () => {
+    const description = describeFor({ standalone: mcpTool("standalone", () => "") }, [])
+    expect(description).toContain("- standalone (1 tool)")
+    expect(description).toContain("tools.standalone.standalone(")
+  })
+
+  test("describeCatalog carries the raw MCP schemas for rendering", () => {
+    const description = describeFor(
+      {
+        weather_current: mcpTool(
+          "current",
+          () => "",
+          { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
+          { type: "object", properties: { tempC: { type: "number" } }, required: ["tempC"] },
+        ),
+      },
+      ["weather"],
+    )
+    expect(description).toContain(
+      "tools.weather.current(input: {\n  city: string,\n}): Promise<{\n  tempC: number,\n}>",
+    )
+  })
+
+  test("the static base description carries no catalog; the registry appends it", async () => {
+    const tool = await build({ github_list_issues: mcpTool("list_issues", () => "") })
+    expect(tool.id).toBe(CODE_MODE_TOOL)
+    expect(tool.description).toBe("Run a confined orchestration script with access to connected MCP tools.")
+    expect(tool.description).not.toContain("Available tools")
+    expect(tool.description).not.toContain("list_issues")
+  })
+
+  test("small catalogs inline every full signature in the appended catalog", () => {
+    const description = describeFor({
+      github_create_issue: mcpTool("create_issue", () => "", {
+        type: "object",
+        properties: { title: { type: "string" }, body: { type: "string" } },
+        required: ["title"],
+      }),
+      github_list_issues: mcpTool("list_issues", () => ""),
+      linear_search: mcpTool("search", () => ""),
+    })
+
+    expect(description).toContain("Available tools (COMPLETE list")
+    expect(description).toContain("- github (2 tools)")
+    expect(description).toContain("- linear (1 tool)")
+    expect(description).toContain(
+      "tools.github.create_issue(input: {\n  title: string,\n  body?: string,\n}): Promise<unknown>",
+    )
+    expect(description).toContain("tools.github.list_issues(")
+    expect(description).toContain("tools.linear.search(")
+    expect(description).toContain("tools.linear.search(input: {}): Promise<unknown>")
+    expect(description).not.toContain("$codemode")
+    expect(description).not.toContain("Browse one namespace")
+    expect(description).toContain("## Workflow")
+    expect(description).toContain("1. Pick a tool from the list under `## Available tools`")
+    expect(description).not.toContain("JSON.parse(res)")
+    expect(description).toContain("check that it is a non-null object and not an array")
+    expect(description).toContain("Return only the fields you need")
+    expect(description).not.toContain("total_count")
+  })
+
+  test("signatures render the declared outputSchema as the return type", () => {
+    const description = describeFor({
+      weather_current: mcpTool(
+        "current",
+        () => "",
+        { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
+        {
+          type: "object",
+          properties: { tempC: { type: "number" }, summary: { type: "string" } },
+          required: ["tempC"],
+        },
+      ),
+    })
+    expect(description).toContain(
+      "tools.weather.current(input: {\n  city: string,\n}): Promise<{\n  tempC: number,\n  summary?: string,\n}>",
+    )
+  })
+
+  test("large catalogs inline a budgeted PARTIAL list plus runtime search", async () => {
+    const tools: Record<string, MCP.McpTool> = {}
+    const filler = "a searchable description of this operation that consumes catalog budget ".repeat(3)
+    for (let i = 0; i < 150; i++) {
+      tools[`alpha_op_${i}`] = {
+        def: {
+          name: `op_${i}`,
+          description: `${filler}${i}`,
+          inputSchema: { type: "object", properties: { value: { type: "string" }, count: { type: "number" } } },
+        } as MCPToolDef,
+        client: { callTool: async () => ({ content: [] }) } as unknown as MCP.McpTool["client"],
+      }
+    }
+    tools["zeta_only_tool"] = mcpTool("only_tool", () => "", {
+      type: "object",
+      properties: { topic: { type: "string", description: "Subject to look up" } },
+      required: ["topic"],
+    })
+    const description = describeFor(tools, ["alpha", "zeta"])
+
+    expect(description).toContain("Available tools (PARTIAL - ")
+    expect(description).toMatch(/- alpha \(150 tools, \d+ shown\)/)
+    expect(description).toContain("- zeta (1 tool)\n")
+    expect(description).toContain(
+      "tools.zeta.only_tool(input: {\n  /** Subject to look up */\n  topic: string,\n}): Promise<unknown>",
+    )
+    expect(description).toContain("tools.$codemode.search(")
+    expect(description).toContain("  limit?: number,\n  offset?: number,")
+    expect(description).toContain("  remaining: number,\n  next: {")
+    expect(description).toContain("      offset: number,\n    } | null,")
+    expect(description).toContain(
+      '1. If needed, discover tools: `return await tools.$codemode.search({ query: "<intent + key nouns>" })`.',
+    )
+    expect(description).toContain(
+      '- Browse one namespace: `await tools.$codemode.search({ query: "", namespace: "<name>" })`.',
+    )
+    expect(description).not.toContain("total_count")
+    expect(description).toContain("tools.alpha.op_0(")
+    expect(description).not.toContain("tools.alpha.op_99(")
+
+    const tool = await build(tools, ["alpha", "zeta"])
+    const out = await Effect.runPromise(
+      tool.execute({ code: "return await tools.$codemode.search({ query: 'only tool', limit: 3, offset: 0 })" }, ctx),
+    )
+    const result = JSON.parse(out.output)
+    expect(result.items.map((i: any) => i.path)).toContain("tools.zeta.only_tool")
+    expect(result).toMatchObject({ remaining: 0, next: null })
+    expect(result.items[0].signature).toContain("tools.")
+    const signature = result.items.find((i: any) => i.path === "tools.zeta.only_tool").signature
+    expect(signature).toContain("tools.zeta.only_tool(input: {\n")
+    expect(signature).toContain("  /** Subject to look up */\n  topic: string")
+    expect(description).toContain("/** Subject to look up */")
+    expect(out.metadata.toolCalls).toEqual([
+      { tool: "$codemode.search", status: "completed", input: { query: "only tool", limit: 3, offset: 0 } },
+    ])
+  })
+
+  test("runs plain JavaScript and returns the value as text", async () => {
+    const tool = await build({})
+    const output = await Effect.runPromise(tool.execute({ code: "return 1 + 2" }, ctx))
+    expect(output.output).toBe("3")
+    expect(output.metadata.toolCalls).toEqual([])
+  })
+
+  test("Object.keys(tools) enumerates the MCP server and CodeMode namespaces", async () => {
+    const tool = await build({
+      github_list_issues: mcpTool("list_issues", () => ""),
+      linear_search: mcpTool("search", () => ""),
+    })
+    const output = await Effect.runPromise(
+      tool.execute(
+        { code: "const namespaces = Object.keys(tools); return { namespaces, count: namespaces.length }" },
+        ctx,
+      ),
+    )
+    expect(JSON.parse(output.output)).toEqual({ namespaces: ["github", "linear", "$codemode"], count: 3 })
+  })
+
+  test("calls a namespaced MCP tool and flows its text result back into the program", async () => {
+    const seen: Record<string, unknown>[] = []
+    const tool = await build({
+      greeter_hello: mcpTool("hello", (args) => {
+        seen.push(args)
+        return { content: [{ type: "text", text: `hello ${args.name}` }] }
+      }),
+    })
+
+    const output = await Effect.runPromise(
+      tool.execute({ code: "const r = await tools.greeter.hello({ name: 'world' }); return r.toUpperCase()" }, ctx),
+    )
+
+    expect(seen).toEqual([{ name: "world" }])
+    expect(output.output).toBe("HELLO WORLD")
+    expect(output.metadata.toolCalls).toEqual([
+      { tool: "greeter.hello", status: "completed", input: { name: "world" } },
+    ])
+  })
+
+  test("exposes structured content as native data and composes multiple calls", async () => {
+    const tool = await build({
+      math_add: mcpTool("add", (args) => ({
+        content: [],
+        structuredContent: { sum: (args.a as number) + (args.b as number) },
+      })),
+    })
+
+    const output = await Effect.runPromise(
+      tool.execute(
+        {
+          code: `
+            const first = await tools.math.add({ a: 1, b: 2 })
+            const second = await tools.math.add({ a: first.sum, b: 10 })
+            return { total: second.sum }
+          `,
+        },
+        ctx,
+      ),
+    )
+
+    expect(JSON.parse(output.output)).toEqual({ total: 13 })
+    expect(output.metadata.toolCalls).toEqual([
+      { tool: "math.add", status: "completed", input: { a: 1, b: 2 } },
+      { tool: "math.add", status: "completed", input: { a: 3, b: 10 } },
+    ])
+  })
+
+  test("runs tool calls in parallel with Promise.all", async () => {
+    const tool = await build({
+      echo_one: mcpTool("one", () => ({ content: [{ type: "text", text: "1" }] })),
+      echo_two: mcpTool("two", () => ({ content: [{ type: "text", text: "2" }] })),
+    })
+
+    const output = await Effect.runPromise(
+      tool.execute(
+        { code: "const [a, b] = await Promise.all([tools.echo.one({}), tools.echo.two({})]); return a + b" },
+        ctx,
+      ),
+    )
+
+    expect(output.output).toBe("12")
+    expect(output.metadata.toolCalls.map((c) => c.tool).sort()).toEqual(["echo.one", "echo.two"])
+    expect(output.metadata.toolCalls.every((c) => c.status === "completed")).toBe(true)
+  })
+
+  test("a program failure fails the tool with a readable error", async () => {
+    const tool = await build({})
+    const error = await failure(tool.execute({ code: "throw new Error('boom')" }, ctx))
+    expect(error.message).toBe("Uncaught: boom")
+  })
+
+  test("reports an unknown tool as a failed execution", async () => {
+    const tool = await build({ known_tool: mcpTool("tool", () => "ok") })
+    const error = await failure(tool.execute({ code: "return await tools.known.missing({})" }, ctx))
+    expect(error.message).toContain("Unknown tool 'known.missing'")
+  })
+
+  test("propagates an MCP tool error into the program as a catchable failure", async () => {
+    const tool = await build({
+      bad_tool: mcpTool("tool", () => ({ isError: true, content: [{ type: "text", text: "server exploded" }] })),
+    })
+    const output = await Effect.runPromise(
+      tool.execute({ code: "try { await tools.bad.tool({}) } catch (e) { return 'caught: ' + e.message }" }, ctx),
+    )
+    expect(output.output).toBe("caught: server exploded")
+  })
+
+  test("asks permission before each child tool call", async () => {
+    const asked: unknown[] = []
+    const permissionCtx: Tool.Context = { ...ctx, ask: (req) => Effect.sync(() => void asked.push(req)) }
+    const ok = () => ({ content: [{ type: "text", text: "ok" }] })
+    const tool = await build({ a_tool: mcpTool("a", ok), b_tool: mcpTool("b", ok) })
+
+    await Effect.runPromise(
+      tool.execute({ code: "await tools.a.tool({}); await tools.b.tool({}); return 'done'" }, permissionCtx),
+    )
+
+    expect(asked.map((req: any) => req.permission)).toEqual(["a_tool", "b_tool"])
+  })
+
+  test("a denied permission fails the child call with a catchable message, not the whole execute", async () => {
+    const denyCtx: Tool.Context = { ...ctx, ask: () => Effect.die(new Error("permission denied by user")) }
+    const called: string[] = []
+    const tool = await build({
+      a_tool: mcpTool("a", () => {
+        called.push("a")
+        return { content: [{ type: "text", text: "ok" }] }
+      }),
+    })
+
+    const output = await Effect.runPromise(
+      tool.execute({ code: "try { await tools.a.tool({}) } catch (e) { return 'denied: ' + e.message }" }, denyCtx),
+    )
+
+    expect(output.output).toBe("denied: permission denied by user")
+    expect(output.metadata.error).toBeUndefined()
+    expect(called).toEqual([])
+    expect(output.metadata.toolCalls).toEqual([{ tool: "a.tool", status: "error" }])
+  })
+
+  test("child calls fire plugin tool.execute hooks with the MCP key and synthetic parent/N call ids", async () => {
+    const events: { name: string; input: any; output: any }[] = []
+    const trigger = ((name: unknown, input: unknown, output: unknown) =>
+      Effect.sync(() => {
+        events.push({ name: name as string, input, output })
+        return output
+      })) as Plugin.Interface["trigger"]
+    const tool = await build(
+      {
+        a_tool: mcpTool("a", () => ({ content: [{ type: "text", text: "one" }] })),
+        b_tool: mcpTool("b", () => ({ content: [{ type: "text", text: "two" }] })),
+      },
+      undefined,
+      undefined,
+      trigger,
+    )
+
+    const out = await Effect.runPromise(
+      tool.execute({ code: "await tools.a.tool({ x: 1 }); await tools.b.tool({}); return 'done'" }, ctx),
+    )
+
+    expect(out.output).toBe("done")
+    expect(events.map((e) => [e.name, e.input.tool, e.input.callID])).toEqual([
+      ["tool.execute.before", "a_tool", "call_code_mode/1"],
+      ["tool.execute.after", "a_tool", "call_code_mode/1"],
+      ["tool.execute.before", "b_tool", "call_code_mode/2"],
+      ["tool.execute.after", "b_tool", "call_code_mode/2"],
+    ])
+    const [before, after] = events
+    expect(before!.input.sessionID).toBe(ctx.sessionID)
+    expect(before!.output).toEqual({ args: { x: 1 } })
+    expect(after!.input.args).toEqual({ x: 1 })
+    expect(after!.output).toEqual({ content: [{ type: "text", text: "one" }] })
+  })
+
+  test("a failing before hook fails only that child call as a catchable in-program error", async () => {
+    const trigger = ((name: unknown, input: any, output: unknown) => {
+      if (name === "tool.execute.before" && input.tool === "a_tool") return Effect.die(new Error("hook exploded"))
+      return Effect.succeed(output)
+    }) as Plugin.Interface["trigger"]
+    const called: string[] = []
+    const record = (name: string) => () => {
+      called.push(name)
+      return { content: [{ type: "text", text: "ok" }] }
+    }
+    const tool = await build(
+      { a_tool: mcpTool("a", record("a")), b_tool: mcpTool("b", record("b")) },
+      undefined,
+      undefined,
+      trigger,
+    )
+
+    const out = await Effect.runPromise(
+      tool.execute(
+        {
+          code: `
+            let caught
+            try { await tools.a.tool({}) } catch (e) { caught = e.message }
+            const r = await tools.b.tool({})
+            return caught + " / " + r
+          `,
+        },
+        ctx,
+      ),
+    )
+
+    expect(out.metadata.error).toBeUndefined()
+    expect(out.output).toBe("hook exploded / ok")
+    expect(called).toEqual(["b"])
+  })
+
+  test("streams live per-call metadata as a call starts and finishes", async () => {
+    const snapshots: Array<{ toolCalls: { tool: string; status: string; input?: Record<string, unknown> }[] }> = []
+    const recordingCtx: Tool.Context = {
+      ...ctx,
+      metadata: (val: any) => Effect.sync(() => void snapshots.push(val.metadata)),
+    }
+    const tool = await build({ greeter_hello: mcpTool("hello", () => ({ content: [{ type: "text", text: "hi" }] })) })
+
+    await Effect.runPromise(
+      tool.execute({ code: "await tools.greeter.hello({ name: 'Ada' }); return 'done'" }, recordingCtx),
+    )
+
+    expect(snapshots).toContainEqual({
+      toolCalls: [{ tool: "greeter.hello", status: "running", input: { name: "Ada" } }],
+    })
+    expect(snapshots).toContainEqual({
+      toolCalls: [{ tool: "greeter.hello", status: "completed", input: { name: "Ada" } }],
+    })
+  })
+
+  test("marks a failed child call as error in the live metadata", async () => {
+    const snapshots: Array<{ toolCalls: { tool: string; status: string; input?: Record<string, unknown> }[] }> = []
+    const recordingCtx: Tool.Context = {
+      ...ctx,
+      metadata: (val: any) => Effect.sync(() => void snapshots.push(val.metadata)),
+    }
+    const tool = await build({
+      bad_tool: mcpTool("tool", () => ({ isError: true, content: [{ type: "text", text: "boom" }] })),
+    })
+
+    await Effect.runPromise(
+      tool.execute(
+        { code: "try { await tools.bad.tool({ reason: 'test' }) } catch (e) { return 'caught' }" },
+        recordingCtx,
+      ),
+    )
+
+    expect(snapshots).toContainEqual({ toolCalls: [{ tool: "bad.tool", status: "error", input: { reason: "test" } }] })
+  })
+
+  test("accumulates stripped media as execute attachments the sandbox never sees", async () => {
+    const tool = await build({
+      shot_take: mcpTool("take", () => ({
+        content: [{ type: "image", data: "PNGDATA", mimeType: "image/png" }],
+        structuredContent: { name: "shot.png" },
+      })),
+    })
+
+    const out = await Effect.runPromise(tool.execute({ code: "return await tools.shot.take({})" }, ctx))
+    expect(JSON.parse(out.output)).toEqual({ name: "shot.png" })
+    expect(out.attachments).toEqual([{ type: "file", mime: "image/png", url: "data:image/png;base64,PNGDATA" }])
+    expect(out.output).not.toContain("PNGDATA")
+  })
+
+  test("a media-only result returns a text marker so the program knows it succeeded", async () => {
+    const tool = await build({
+      shot_take: mcpTool("take", () => ({ content: [{ type: "image", data: "PNGDATA", mimeType: "image/png" }] })),
+    })
+    const out = await Effect.runPromise(tool.execute({ code: "return await tools.shot.take({})" }, ctx))
+    expect(out.output).toBe("[1 image attached to the result]")
+    expect(out.attachments).toEqual([{ type: "file", mime: "image/png", url: "data:image/png;base64,PNGDATA" }])
+  })
+
+  test("media-only markers distinguish all-image from mixed attachments", async () => {
+    const tool = await build({
+      media_images: mcpTool("images", () => ({
+        content: [
+          { type: "image", data: "PNG1", mimeType: "image/png" },
+          { type: "image", data: "PNG2", mimeType: "image/png" },
+        ],
+      })),
+      media_mixed: mcpTool("mixed", () => ({
+        content: [
+          { type: "image", data: "PNG3", mimeType: "image/png" },
+          { type: "resource", resource: { uri: "file:///tmp/report.pdf", mimeType: "application/pdf", blob: "PDF1" } },
+        ],
+      })),
+    })
+    const out = await Effect.runPromise(
+      tool.execute(
+        {
+          code: `
+            const images = await tools.media.images({})
+            const mixed = await tools.media.mixed({})
+            return { images, mixed }
+          `,
+        },
+        ctx,
+      ),
+    )
+
+    expect(JSON.parse(out.output)).toEqual({
+      images: "[2 images attached to the result]",
+      mixed: "[2 files attached to the result]",
+    })
+    expect(out.output).not.toContain("PNG")
+    expect(out.attachments).toEqual([
+      { type: "file", mime: "image/png", url: "data:image/png;base64,PNG1" },
+      { type: "file", mime: "image/png", url: "data:image/png;base64,PNG2" },
+      { type: "file", mime: "image/png", url: "data:image/png;base64,PNG3" },
+      { type: "file", mime: "application/pdf", url: "data:application/pdf;base64,PDF1", filename: "report.pdf" },
+    ])
+  })
+
+  test("resource links flow to the program as text, never as attachments", async () => {
+    const tool = await build({
+      docs_find: mcpTool("find", () => ({
+        content: [
+          {
+            type: "resource_link",
+            uri: "https://example.com/guide.pdf",
+            name: "guide.pdf",
+            mimeType: "application/pdf",
+          },
+          { type: "resource_link", uri: "file:///tmp/notes.md", name: "notes.md" },
+        ],
+      })),
+    })
+    const out = await Effect.runPromise(tool.execute({ code: "return await tools.docs.find({})" }, ctx))
+
+    expect(out.output).toBe("guide.pdf: https://example.com/guide.pdf\nnotes.md: file:///tmp/notes.md")
+    expect(out.attachments).toBeUndefined()
+  })
+
+  test("attachments still flow when the program returns something else entirely", async () => {
+    const tool = await build({
+      shot_take: mcpTool("take", () => ({ content: [{ type: "image", data: "PNGDATA", mimeType: "image/png" }] })),
+    })
+    const out = await Effect.runPromise(tool.execute({ code: "await tools.shot.take({}); return 'captured'" }, ctx))
+    expect(out.output).toBe("captured")
+    expect(out.attachments).toHaveLength(1)
+  })
+
+  test("isolates the sandbox from host globals", async () => {
+    const tool = await build({})
+    const error = await failure(tool.execute({ code: "return process.env" }, ctx))
+    expect(error.message).toContain("process")
+  })
+
+  test("cancelling via ctx.abort interrupts the running program", async () => {
+    const controller = new AbortController()
+    const tool = await build({
+      host_trigger: mcpTool("trigger", () => {
+        controller.abort()
+        return new Promise(() => {})
+      }),
+    })
+    const output = await Effect.runPromise(
+      tool.execute(
+        { code: "try { await tools.host.trigger({}) } catch {} while (true) {}" },
+        { ...ctx, abort: controller.signal },
+      ),
+    )
+    expect(output.output).toBe("Execution cancelled.")
+    expect(output.metadata.error).toBe(true)
+    expect(output.metadata.toolCalls).toEqual([{ tool: "host.trigger", status: "running" }])
+  })
+
+  test("a pre-aborted signal cancels before the program runs", async () => {
+    const controller = new AbortController()
+    controller.abort()
+    const ran: string[] = []
+    const tool = await build({ host_touch: mcpTool("touch", () => (ran.push("called"), "ok")) })
+    const output = await Effect.runPromise(
+      tool.execute({ code: "return await tools.host.touch({})" }, { ...ctx, abort: controller.signal }),
+    )
+    expect(output.output).toBe("Execution cancelled.")
+    expect(ran).toEqual([])
+  })
+
+  test("leaves oversized results to OpenCode's native tool-output truncation", async () => {
+    const tool = await build({})
+    const output = await Effect.runPromise(tool.execute({ code: "return 'x'.repeat(40000)" }, ctx))
+    expect(output.metadata.error).toBeUndefined()
+    expect(output.output).not.toContain("[result truncated:")
+    expect(output.output.length).toBeGreaterThanOrEqual(40_000)
+  })
+
+  test("appends logs after the result on success and after the message on error", async () => {
+    const tool = await build({})
+
+    const ok = await Effect.runPromise(
+      tool.execute({ code: "console.log('step one'); console.warn('careful'); return 'done'" }, ctx),
+    )
+    expect(ok.output).toBe("done\n\nLogs:\nstep one\n[warn] careful")
+
+    const error = await failure(tool.execute({ code: "console.log('before the throw'); throw new Error('boom')" }, ctx))
+    expect(error.message).toContain("Uncaught: boom")
+    expect(error.message).toContain("Logs:\nbefore the throw")
+  })
+})
+
+describe("code mode permission visibility", () => {
+  const deny = (permission: string): PermissionV1.Rule => ({ permission, pattern: "*", action: "deny" })
+  const askRule = (permission: string): PermissionV1.Rule => ({ permission, pattern: "*", action: "ask" })
+  const ok = () => ({ content: [{ type: "text", text: "ok" }] })
+
+  test("a hard-denied tool never enters the catalog or its search index", () => {
+    const mcpTools = {
+      github_create_issue: mcpTool("create_issue", ok),
+      github_list_issues: mcpTool("list_issues", ok),
+    }
+    const description = describeFor(mcpTools, ["github"], [deny("github_create_issue")])
+    expect(description).toContain("tools.github.list_issues(")
+    expect(description).not.toContain("create_issue")
+    expect(description).toContain("- github (1 tool)")
+  })
+
+  test("an ask-level tool stays fully visible in the catalog", () => {
+    const mcpTools = {
+      github_create_issue: mcpTool("create_issue", ok),
+      github_list_issues: mcpTool("list_issues", ok),
+    }
+    const description = describeFor(mcpTools, ["github"], [askRule("github_create_issue")])
+    expect(description).toContain("tools.github.create_issue(")
+    expect(description).toContain("tools.github.list_issues(")
+    expect(description).toContain("- github (2 tools)")
+  })
+
+  test("a hard-denied tool is not dispatchable: the program gets the unknown-tool diagnostic", async () => {
+    const called: string[] = []
+    const tool = await build(
+      {
+        github_create_issue: mcpTool("create_issue", () => {
+          called.push("create_issue")
+          return ok()
+        }),
+        github_list_issues: mcpTool("list_issues", ok),
+      },
+      ["github"],
+      [deny("github_create_issue")],
+    )
+
+    const denied = await failure(tool.execute({ code: "return await tools.github.create_issue({ title: 'x' })" }, ctx))
+    expect(denied.message).toContain("Unknown tool 'github.create_issue'")
+    expect(denied.message).not.toContain("permission")
+    expect(called).toEqual([])
+
+    const allowed = await Effect.runPromise(tool.execute({ code: "return await tools.github.list_issues({})" }, ctx))
+    expect(allowed.metadata.error).toBeUndefined()
+    expect(allowed.output).toBe("ok")
+  })
+
+  test("an ask-level tool remains callable and still prompts via ctx.ask", async () => {
+    const asked: string[] = []
+    const askCtx: Tool.Context = { ...ctx, ask: (req) => Effect.sync(() => void asked.push(req.permission)) }
+    const tool = await build(
+      { github_list_issues: mcpTool("list_issues", ok) },
+      ["github"],
+      [askRule("github_list_issues")],
+    )
+    const out = await Effect.runPromise(tool.execute({ code: "return await tools.github.list_issues({})" }, askCtx))
+    expect(out.output).toBe("ok")
+    expect(asked).toEqual(["github_list_issues"])
+  })
+
+  test("Permission.visibleTools hides only hard denies, matching Permission.disabled", () => {
+    const tools = { a_tool: 1, b_tool: 2, c_tool: 3 }
+    const visible = Permission.visibleTools(tools, [
+      deny("a_tool"),
+      askRule("b_tool"),
+      { permission: "c_tool", pattern: "something", action: "deny" },
+    ])
+    expect(Object.keys(visible)).toEqual(["b_tool", "c_tool"])
+  })
+})

packages/opencode/test/tool/registry.test.ts

@@ -19,6 +19,8 @@ import { MessageID, SessionID } from "@/session/schema"
 import { RuntimeFlags } from "@/effect/runtime-flags"
 import { ProviderV2 } from "@opencode-ai/core/provider"
 import { ModelV2 } from "@opencode-ai/core/model"
+import { MCP } from "@/mcp"
+import type { Tool as MCPToolDef } from "@modelcontextprotocol/sdk/types.js"
 
 const configLayer = TestConfig.layer({
   directories: () => InstanceState.directory.pipe(Effect.map((dir) => [path.join(dir, ".opencode")])),
@@ -55,6 +57,42 @@ const replacements = [
 ] as const
 
 const it = testEffect(LayerNode.compile(root, replacements))
+const withCodeMode = testEffect(
+  LayerNode.compile(root, [
+    [Config.node, configLayer],
+    [RuntimeFlags.node, RuntimeFlags.layer({ experimentalCodeMode: true })],
+    [
+      MCP.node,
+      Layer.mock(MCP.Service, {
+        tools: () =>
+          Effect.succeed({
+            weather_current: {
+              def: {
+                name: "current",
+                description: "current weather",
+                inputSchema: { type: "object", properties: { city: { type: "string" } }, required: ["city"] },
+              } as MCPToolDef,
+              client: {} as MCP.McpTool["client"],
+            },
+          }),
+        clients: () => Effect.succeed({ weather: {} as any }),
+      }),
+    ],
+  ]),
+)
+const withEmptyCodeMode = testEffect(
+  LayerNode.compile(root, [
+    [Config.node, configLayer],
+    [RuntimeFlags.node, RuntimeFlags.layer({ experimentalCodeMode: true })],
+    [
+      MCP.node,
+      Layer.mock(MCP.Service, {
+        tools: () => Effect.succeed({}),
+        clients: () => Effect.succeed({}),
+      }),
+    ],
+  ]),
+)
 const withBrokenPlugin = testEffect(LayerNode.compile(root, [...replacements, [Plugin.node, brokenPluginLayer]]))
 
 afterEach(async () => {
@@ -71,6 +109,47 @@ describe("tool.registry", () => {
     }),
   )
 
+  it.instance("does not expose execute unless code mode is enabled", () =>
+    Effect.gen(function* () {
+      const registry = yield* ToolRegistry.Service
+      const ids = yield* registry.ids()
+
+      expect(ids).not.toContain("execute")
+    }),
+  )
+
+  withCodeMode.instance("exposes execute when code mode is enabled", () =>
+    Effect.gen(function* () {
+      const registry = yield* ToolRegistry.Service
+      const agents = yield* Agent.Service
+      const ids = yield* registry.ids()
+      const tools = yield* registry.tools({
+        providerID: ProviderV2.ID.opencode,
+        modelID: ModelV2.ID.make("test"),
+        agent: yield* agents.defaultInfo(),
+      })
+      const execute = tools.find((tool) => tool.id === "execute")
+
+      expect(ids).toContain("execute")
+      expect(tools.map((tool) => tool.id)).toContain("execute")
+      expect(execute?.description).toContain("tools.weather.current(input: {\n  city: string,\n})")
+    }),
+  )
+
+  withEmptyCodeMode.instance("does not expose execute when code mode has no visible tools", () =>
+    Effect.gen(function* () {
+      const registry = yield* ToolRegistry.Service
+      const agents = yield* Agent.Service
+      const tools = yield* registry.tools({
+        providerID: ProviderV2.ID.opencode,
+        modelID: ModelV2.ID.make("test"),
+        agent: yield* agents.defaultInfo(),
+      })
+
+      expect(tools.map((tool) => tool.id)).not.toContain("execute")
+    }),
+  )
+
   it.instance("hides task background parameter unless experimental background subagents are enabled", () =>
     Effect.gen(function* () {
       const registry = yield* ToolRegistry.Service