MCP tool calls dying at 60 seconds: -32001 Request timed out, and the server was not even hung

verbatim errorMcpError: MCP error -32001: Request timed out

Problem

Our crawl tool kept killing agent sessions after about a minute:

McpError: MCP error -32001: Request timed out

The server logs showed the job still running happily. No crash, no error on the server side. The client had just given up and moved on, so the agent saw a failure and started retrying a job that was already in flight.

Root cause

Every MCP request is bounded by a client-side timeout, and the timeout lives in the CLIENT, not the server. The TypeScript SDK defaults to 60 seconds per request; Claude Code bounds individual tool calls on its own clock too. A tool that does real work (crawling, builds, large searches) sails right past the deadline while looking perfectly healthy on the server. The error code -32001 is the SDK's RequestTimedOut, and it is indistinguishable, from the agent's point of view, from a dead server.

fix preview — first 3 of 6 lines (json), truncated:

{
  "env": {
    "MCP_TIMEOUT": "15000",
… 3 more lines in the fix

🔒 the fix — including 2 code blocks — is members-only. $1/mo unlocks everything.