René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"RequestResponder.__exit__ leaks CancelledError on cancelled request, killing the stdio receive loop","articleBody":"## Summary\n\nWhen a client sends `notifications/cancelled` for an in-flight request handled over stdio, the server's receive loop task group dies. The process stays alive but stops reading stdin, so every subsequent request hangs and the client eventually reports `MCP error -32000: Connection closed`. The bug is racy — it reproduces in roughly 40–80% of attempts depending on platform timing.\n\nAffects: `mcp` 1.26.0 and 1.27.1 (latest at time of writing). Confirmed with FastMCP 3.1.x and the in-tree stdio server.\n\n## Reproduction\n\nA standalone subprocess test is in [mlorentedev/hive `tests/test_transport_recovery.py`](https://github.com/mlorentedev/hive/blob/master/tests/test_transport_recovery.py). The relevant flow:\n\n1. `initialize` → ack\n2. `notifications/initialized`\n3. `tools/call id=2`\n4. `notifications/cancelled` for `requestId=2` (within a few ms of step 3)\n5. Receive: `{\"id\": 2, \"error\": {\"code\": 0, \"message\": \"Request cancelled\"}}` ← OK\n6. `tools/call id=3`\n7. **No response.** Server is alive but `proc.stdout.readline()` hangs.\n\n## Root cause\n\n`mcp.shared.session.RequestResponder.__exit__`:\n\n```python\ndef __exit__(self, exc_type, exc_val, exc_tb):\n    try:\n        if self._completed:\n            self._on_complete(self)\n    finally:\n        self._entered = False\n        ...\n        self._cancel_scope.__exit__(exc_type, exc_val, exc_tb)  # ← (A)\n```\n\nWhen `notifications/cancelled` arrives, `RequestResponder.cancel()` calls `self._cancel_scope.cancel()` and sends the error response. The handler task catches the `CancelledError` in `mcp/server/lowlevel/server.py` (around line 766) and `return`s. The `with responder:` block then exits with `exc_type=None` while the cancel scope is still in cancelled state — at line (A), anyio's `CancelScope.__exit__` re-raises `CancelledError`.\n\nThat exception bubbles up to:\n\n```python\nasync with anyio.create_task_group() as tg:\n    async for message in session.incoming_messages:\n        tg.start_soon(self._handle_message, ...)\n```\n\n…in `Server._run`. anyio task groups cancel all sibling tasks and propagate the cancellation. The receive loop is one of those sibling tasks, so it dies.\n\n## Suggested fix\n\nSwallow the spurious cancellation when the responder has already sent its response:\n\n```python\ndef __exit__(self, exc_type, exc_val, exc_tb):\n    try:\n        if self._completed:\n            self._on_complete(self)\n    finally:\n        self._entered = False\n        if not self._cancel_scope:\n            raise RuntimeError(\"No active cancel scope\")\n        try:\n            self._cancel_scope.__exit__(exc_type, exc_val, exc_tb)\n        except BaseException as exc:\n            if self._completed and isinstance(exc, anyio.get_cancelled_exc_class()):\n                # cancel() already sent the error response — the scope's\n                # re-raised cancellation is spurious.\n                return\n            raise\n```\n\nMirrors what we ship in [hive `src/hive/_compat.py`](https://github.com/mlorentedev/hive/blob/master/src/hive/_compat.py).\n\n## Workaround used downstream\n\nWe monkey-patch `RequestResponder.__exit__` at import time. The patch is self-gated (only fires when `_completed=True` AND the leaking exception is `anyio.get_cancelled_exc_class()`), so it stays inert once a fix lands upstream.\n\n## Environment\n\n- Reproduced on Windows 11 (`mcp` 1.26.0, 1.27.1) and via Claude Code as the host.\n- Tracked downstream in [mlorentedev/hive#75](https://github.com/mlorentedev/hive/issues/75).\n- Regression test passes 5/5 with the patch applied on Python 3.12; fails 2/5 — 4/5 without it.\n- Python 3.13 appears to have additional uncancel semantics that the patch does not yet fully cover — verification in progress.","author":{"url":"https://github.com/mlorentedev","@type":"Person","name":"mlorentedev"},"datePublished":"2026-05-15T19:52:52.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":4},"url":"https://github.com/2610/python-sdk/issues/2610"}