fix(mcp): recover expired legacy sessions

packages/opencode/src/mcp/catalog.ts

@@ -1,4 +1,4 @@
-import { Client, type CallToolResult, type Tool as MCPToolDef } from "@modelcontextprotocol/client"
+import { Client, SdkHttpError, type CallToolResult, type Tool as MCPToolDef } from "@modelcontextprotocol/client"
 import { dynamicTool, jsonSchema, type JSONSchema7, type Tool } from "ai"
 import { Effect } from "effect"
 
@@ -8,6 +8,7 @@ export interface McpTool {
   readonly def: MCPToolDef
   readonly client: Client
   readonly timeout?: number
+  readonly recover?: () => Promise<Client | undefined>
 }
 
 export async function callTool(
@@ -15,16 +16,24 @@ export async function callTool(
   args: Record<string, unknown>,
   signal?: AbortSignal,
 ): Promise<CallToolResult> {
-  const result = await tool.client.callTool(
-    { name: tool.def.name, arguments: args },
-    {
-      resetTimeoutOnProgress: true,
-      signal,
-      timeout: tool.timeout,
-      // The MCP SDK only sends a progress token when this hook is present, enabling timeout resets.
-      onprogress: () => {},
-    },
-  )
+  const invoke = (client: Client) =>
+    client.callTool(
+      { name: tool.def.name, arguments: args },
+      {
+        resetTimeoutOnProgress: true,
+        signal,
+        timeout: tool.timeout,
+        // The MCP SDK only sends a progress token when this hook is present, enabling timeout resets.
+        onprogress: () => {},
+      },
+    )
+  const result = await invoke(tool.client).catch(async (error) => {
+    if (!tool.recover || !(error instanceof SdkHttpError) || error.status !== 404) throw error
+    const client = await tool.recover()
+    if (!client) throw error
+    signal?.throwIfAborted()
+    return invoke(client)
+  })
   if (result.isError)
     throw new Error(
       result.content

packages/opencode/src/mcp/index.ts

@@ -593,8 +593,13 @@ const layer = Layer.effect(
       listed: MCPToolDef[],
       instructions: string | undefined,
       timeout?: number,
+      expected?: MCPClient,
     ) {
       const bridge = yield* EffectBridge.make()
+      if (expected && s.clients[name] !== expected) {
+        yield* Effect.tryPromise(() => client.close()).pipe(Effect.ignore)
+        return s.status[name] ?? { status: "disabled" }
+      }
       const previous = s.clients[name]
       s.status[name] = { status: "connected" }
       s.clients[name] = client
@@ -656,6 +661,37 @@ const layer = Layer.effect(
       return yield* storeClient(s, name, result.mcpClient, result.defs!, result.instructions, mcp.timeout)
     })
 
+    const reconnectStale = Effect.fn("MCP.reconnectStale")(function* (name: string, stale: MCPClient) {
+      const s = yield* InstanceState.get(state)
+      if (s.clients[name] !== stale) return undefined
+      const mcp = yield* getMcpConfig(name)
+      if (!mcp) return undefined
+
+      yield* Effect.logWarning("reconnecting MCP server after stale session", { server: name })
+      const result = yield* create(name, { ...mcp, enabled: true })
+      if (s.clients[name] !== stale) {
+        if (result.mcpClient) yield* Effect.tryPromise(() => result.mcpClient!.close()).pipe(Effect.ignore)
+        return undefined
+      }
+      if (!result.mcpClient) {
+        s.status[name] = result.status
+        yield* closeClient(s, name)
+        return undefined
+      }
+      yield* storeClient(
+        s,
+        name,
+        result.mcpClient,
+        result.defs!,
+        result.instructions,
+        mcp.timeout,
+        stale,
+      )
+      return s.clients[name]
+    })
+
+    const recoveries = new WeakMap<MCPClient, Promise<MCPClient | undefined>>()
+
     const add = Effect.fn("MCP.add")(function* (name: string, mcp: ConfigMCPV1.Info) {
       const s = yield* InstanceState.get(state)
       s.config[name] = mcp
@@ -684,6 +720,7 @@ const layer = Layer.effect(
     const tools = Effect.fn("MCP.tools")(function* () {
       const result: Record<string, McpTool> = {}
       const s = yield* InstanceState.get(state)
+      const bridge = yield* EffectBridge.make()
 
       const cfg = yield* cfgSvc.get()
       const config = cfg.mcp ?? {}
@@ -698,8 +735,19 @@ const layer = Layer.effect(
           continue
         }
         const timeout = requestTimeout(s, clientName, mcpConfig, defaultTimeout)
+        const transport = client.transport
+        const recover =
+          transport instanceof StreamableHTTPClientTransport && transport.sessionId && client.getProtocolEra() === "legacy"
+            ? () => {
+                const current = recoveries.get(client)
+                if (current) return current
+                const created = bridge.promise(reconnectStale(clientName, client))
+                recoveries.set(client, created)
+                return created
+              }
+            : undefined
         for (const def of listed) {
-          result[McpCatalog.toolName(clientName, def.name)] = { def, client, timeout }
+          result[McpCatalog.toolName(clientName, def.name)] = { def, client, timeout, recover }
         }
       }
       return result

packages/opencode/test/mcp/catalog.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, test } from "bun:test"
-import { Client, InMemoryTransport } from "@modelcontextprotocol/client"
+import { Client, InMemoryTransport, SdkErrorCode, SdkHttpError } from "@modelcontextprotocol/client"
 import { Server } from "@modelcontextprotocol/server"
 import { McpCatalog } from "@/mcp/catalog"
 import { Effect } from "effect"
@@ -86,6 +86,51 @@ describe("McpCatalog.callTool", () => {
 
     await expect(McpCatalog.callTool({ def: mcpTool(), client }, {})).rejects.toThrow("first\n\nsecond")
   })
+
+  test("retries a stale session only once", async () => {
+    const error = new SdkHttpError(SdkErrorCode.SendFailed, "expired", { status: 404 })
+    const client = { callTool: () => Promise.reject(error) } as unknown as Client
+    let recoveries = 0
+
+    await expect(
+      McpCatalog.callTool(
+        {
+          def: mcpTool(),
+          client,
+          recover: async () => {
+            recoveries++
+            return client
+          },
+        },
+        {},
+      ),
+    ).rejects.toBe(error)
+    expect(recoveries).toBe(1)
+  })
+
+  test("does not replay after cancellation during recovery", async () => {
+    const error = new SdkHttpError(SdkErrorCode.SendFailed, "expired", { status: 404 })
+    const controller = new AbortController()
+    let calls = 0
+    const stale = { callTool: () => Promise.reject(error) } as unknown as Client
+    const fresh = { callTool: async () => ({ content: [], calls: ++calls }) } as unknown as Client
+
+    await expect(
+      McpCatalog.callTool(
+        {
+          def: mcpTool(),
+          client: stale,
+          recover: async () => {
+            controller.abort()
+            return fresh
+          },
+        },
+        {},
+        controller.signal,
+      ),
+    ).rejects.toBeInstanceOf(DOMException)
+    expect(calls).toBe(0)
+  })
 })
 
 test("preserves output schema validation across paginated tool discovery", async () => {

packages/opencode/test/mcp/lifecycle.test.ts

@@ -11,6 +11,7 @@ import { LayerNode } from "@opencode-ai/core/effect/layer-node"
 import { Cause, Effect, Exit } from "effect"
 import type { MCP as MCPNS } from "../../src/mcp/index"
 import { MCP } from "../../src/mcp/index"
+import { McpCatalog } from "../../src/mcp/catalog"
 import { McpOAuthCallback } from "../../src/mcp/oauth-callback"
 import { TestInstance } from "../fixture/fixture"
 import { pollWithTimeout, testEffect } from "../lib/effect"
@@ -36,7 +37,12 @@ interface LifecycleServerState {
   aborted: number
 }
 
-function lifecycleServer(input?: { capabilities?: ServerCapabilities; instructions?: string; requestRoots?: boolean }) {
+function lifecycleServer(input?: {
+  capabilities?: ServerCapabilities
+  instructions?: string
+  requestRoots?: boolean
+  legacy?: boolean
+}) {
   const capabilities = input?.capabilities ?? { tools: {}, prompts: {}, resources: {} }
   return Effect.acquireRelease(
     Effect.promise(async () => {
@@ -52,7 +58,11 @@ function lifecycleServer(input?: { capabilities?: ServerCapabilities; instructio
       const makeProtocol = async () => {
         const protocol = new Server(
           { name: "mcp-lifecycle", version: "1.0.0" },
-          { capabilities, instructions: input?.instructions },
+          {
+            capabilities,
+            instructions: input?.instructions,
+            supportedProtocolVersions: input?.legacy ? ["2025-11-25"] : undefined,
+          },
         )
         const transport = new WebStandardStreamableHTTPServerTransport({
           sessionIdGenerator: () => crypto.randomUUID(),
@@ -65,6 +75,9 @@ function lifecycleServer(input?: { capabilities?: ServerCapabilities; instructio
             const page = state.toolPages?.[request.params?.cursor ?? "initial"]
             return Promise.resolve({ tools: page?.items ?? state.tools, nextCursor: page?.nextCursor })
           })
+          protocol.setRequestHandler("tools/call", (request) =>
+            Promise.resolve({ content: [{ type: "text", text: request.params.name }] }),
+          )
         }
         if (capabilities.prompts) {
           protocol.setRequestHandler("prompts/list", (request) => {
@@ -114,6 +127,10 @@ function lifecycleServer(input?: { capabilities?: ServerCapabilities; instructio
         fetch(request) {
           state.requests.push(request.method)
           request.signal.addEventListener("abort", () => state.aborted++)
+          const session = request.headers.get("mcp-session-id")
+          if (request.method === "POST" && session && session !== current.transport.sessionId) {
+            return new Response("Session not found", { status: 404 })
+          }
           return current.transport.handleRequest(request)
         },
       })
@@ -322,6 +339,28 @@ it.instance("disconnect removes protocol data and reconnect establishes a new se
   }),
 )
 
+it.instance("recovers a legacy HTTP session after the server discards it", () =>
+  Effect.gen(function* () {
+    const server = yield* lifecycleServer({ legacy: true })
+    const mcp = yield* MCP.Service
+    yield* mcp.add("session-server", remote(server.url))
+    const tool = (yield* mcp.tools())["session-server_test_tool"]
+    const client = tool?.client
+    expect(tool).toBeDefined()
+
+    yield* Effect.promise(server.restart)
+    const results = yield* Effect.promise(() =>
+      Promise.all([McpCatalog.callTool(tool!, {}), McpCatalog.callTool(tool!, {})]),
+    )
+
+    expect(results.map((result) => result.content)).toEqual([
+      [{ type: "text", text: "test_tool" }],
+      [{ type: "text", text: "test_tool" }],
+    ])
+    expect((yield* mcp.clients())["session-server"]).not.toBe(client)
+  }),
+)
+
 it.instance("add() closes the old protocol session when replacing a server", () =>
   Effect.gen(function* () {
     const first = yield* lifecycleServer()