AgentFixboard
OpenLast checked MCP Python SDK#error-messages#transport

A stateful Streamable HTTP session crash returns an empty -32603

When app.run() raises inside a stateful Streamable HTTP session, the client gets JSON-RPC -32603 with message Error handling POST request: (empty after the colon). The real exception is only in a separate Session <id> crashed log.

From GitHub

modelcontextprotocol/python-sdk

Issue

Difficulty
Easy
Time
~5 min
Steps
2
Fixed in
—
Last checked

Not fixed upstream yet: the issue is open. Source: issue open. Last checked 2026-10-19.

Symptom

When app.run() raises inside a stateful Streamable HTTP session, the client gets JSON-RPC -32603 with message Error handling POST request: (empty after the colon). The real exception is only in a separate Session <id> crashed log.

Evidence what you'll see

  • The client body is {"jsonrpc":"2.0","id":"server-error","error":{"code":-32603,"message":"Error handling POST request: "}} with HTTP 500.
  • The server log shows Session <id> crashed then RuntimeError: BOOM-distinctive-root-cause, and separately Error handling POST request / anyio.ClosedResourceError.
  • The reporter's live trigger was OSError: [Errno 24] inotify instance limit reached; every later request 500'd with an opaque ClosedResourceError.

Why it happens

_handle_stateful_request starts run_server (app.run() on the session streams) and concurrently calls http_transport.handle_request(). If app.run() raises, connect() tears down the streams. _handle_post_request then raises ClosedResourceError at writer.send(session_message), builds the 500 with f"Error handling POST request: {err}", and str(ClosedResourceError()) is empty.

Affected versions

Python 3.14, mcp 1.27.2 (also observed on 1.27.1), anyio 4.13, starlette 1.2.1. Stateful Streamable HTTP (stateless_http=False).

Fix steps

0 of 2 done

Your check marks stay in this browser only.

How to check it worked

  • The JSON-RPC error for that initialize names BOOM-distinctive-root-cause (or RuntimeError), not Error handling POST request: with an empty suffix.

If it still fails

  • The linked issue does not name an SDK release that puts the session crash cause on the client error.
  • This path is the remaining diagnosability gap after the process-crash work in 1.10.0. Closed pull request 2742 was described as sending the real cause to the client; it did not merge.

Sources and credit

Summarized in our own words from these public reports. Credit goes to their authors.

Provided as is, without warranty. Try changes in a non-production environment and back up first. Security-related steps: verify against the official docs. Terms · Request removal or attribution change

Hit an error we don't cover? Submit it