MCP TS SDK: request times out after 60 seconds even with timeout option or progress updates

Question

Using the TypeScript MCP SDK client on Node v23.10.0, a long-running tool call fails with McpError: MCP error -32001: Request timed out after 60000 ms, even though the server sends progress updates every 5 seconds and the Python SDK client handles the same server fine. Passing a timeout to callTool or connect does not seem to override the default. What actually works?

Originally asked on GitHub: https://github.com/modelcontextprotocol/typescript-sdk/issues/245

AgentArmor team · Accepted Answer

The issue is reported as closed, but no SDK-side fix is confirmed in this thread. Lacking other settings, the TypeScript SDK requests time out after 60 seconds and progress notifications do not reset that timer.

Options that appeared in the thread:

1. Pass request options as the third argument to `callTool`:

```js
await client.callTool(
  { name: "your_tool", arguments: {...} },
  undefined,
  { timeout: 300000 },
);
```

One commenter said this seemed to extend the timeout for them. The original reporter said it did not help on their provided test server, and another user said passing timeout in both `connect` and `callTool` did not override the 60-second default.

2. Set `resetTimeoutOnProgress: true` in the request options. A commenter quoted SDK documentation saying that receiving a progress notification then resets the request timeout, and that the server should send regular progress updates for long operations. No maintainer confirmed this in the thread, and a later commenter reported still timing out even with a one-day timeout and `resetTimeoutOnProgress` set, suspecting the request options were not reaching the protocol layer.

3. If more than one client connects, look at the server architecture. One commenter reproduced the timeout with two concurrent clients and found the root cause: their server created only one `McpServer` instance, and the last client connection stole it, so the other client's tool calls timed out. Creating one `McpServer` per client session/transport fixed it (SDK ^1.13.0, SSE transport via `SSEServerTransport`).

General advice: make sure the server actually sends progress notifications, set `resetTimeoutOnProgress` on the client, and do not share a single `McpServer` instance across transports.

AgentArmor team · Answer

**What a GitHub user tried on GitHub** (Didn't work)

Reporter ran a Node.js client against an SSE MCP server whose tool runs about 125 seconds while sending progress every 5 seconds. The request failed at 60000 ms with McpError -32001, while the equivalent Python client succeeded.

Source: https://github.com/modelcontextprotocol/typescript-sdk/issues/245

AgentArmor team · Answer

**What @jrubins tried on GitHub** (Partly worked)

A commenter passed `{ timeout: 300000 }` as the third argument to `client.callTool`. It seemed to extend the timeout for them, but the original reporter said it did not help with the provided test server.

Source: https://github.com/modelcontextprotocol/typescript-sdk/issues/245#issuecomment-2778983612

AgentArmor team · Answer

**What @AbhishekkGautam tried on GitHub** (Didn't work)

Another user tried passing the timeout value in both `connect` and `callTool`, but it did not override the default 60-second timeout.

Source: https://github.com/modelcontextprotocol/typescript-sdk/issues/245#issuecomment-2802865002

AgentArmor team · Answer

**What @mikestaub tried on GitHub** (Outcome unclear)

A commenter suggested setting `resetTimeoutOnProgress: true` in the request options so that receiving a progress notification resets the request timeout. The reporter of this suggestion gave no confirmation from a maintainer, and later reports were mixed.

Source: https://github.com/modelcontextprotocol/typescript-sdk/issues/245#issuecomment-2902077243

AgentArmor team · Answer

**What @ljdean1 tried on GitHub** (Worked)

A commenter with two concurrent clients traced the timeout to a single shared `McpServer` instance being taken over by the last client connection. Creating one `McpServer` per client session/transport, using SDK ^1.13.0 and SSEServerTransport, resolved it for them.

Source: https://github.com/modelcontextprotocol/typescript-sdk/issues/245#issuecomment-2993382026

MCP TS SDK: request times out after 60 seconds even with timeout option or progress updates

6 answers

Add your answer or comment

MCP TS SDK: request times out after 60 seconds even with timeout option or progress updates

6 answers

Add your answer or comment

Related questions