MCP Python SDK idle session 404: what it means for Sume jobs

The 30-minute idle 404 is a behavior of stateful servers built with the Python SDK, and nothing in the Sume code I read produces it: the hosted endpoint takes POST /mcp calls and dispatches each by method. If a client of yours hits a 404 and re-initializes, your Sume jobs are unaffected. Keep the job ids and read them again.

SDK facts are from the Python SDK 2.2.0 release notes; Sume facts from the server source and Jobs and results, read 2026-09-30.

What does the SDK change?

The notes say a stateful session with nothing in flight for 30 minutes is closed; the client's next request gets a 404 and it must initialize again. They give session_idle_timeout=None, max_sessions=None on mcp.run(transport="streamable-http", ...) to turn the limits off.

What does a Sume client see instead?

How do I wait on a long Sume job?

Each remote HTTP call holds at most 55 seconds (default 50). On wait_slice_expired, call jobs_wait again with the same ids. The docs say never to resubmit the paid create. That loop does not depend on any one connection staying open; see jobs_wait for long video jobs.

What if my own Python MCP server sits in front?

Then the 30-minute rule applies to that server, not to Sume. Either handle the 404 by initializing again, or set the timeout options the notes list. Calls to Sume from inside your tools still need a stable idempotency_key so a retry does not spend twice; see what stateless MCP means.