fix(sse): Handle ClosedResourceError during shutdown

Improved the fix for the cancel scope lifecycle violation by adding
graceful handling of ClosedResourceError. When streams close during
shutdown, the error handler may attempt to send exceptions to already
closed streams, which should be ignored as this is expected behavior.

This maintains the original fix (removing manual tg.cancel_scope.cancel())
while handling the race condition it exposed in error cleanup.

Author: ryanbaumann

src/mcp/client/sse.py

@@ -111,7 +111,11 @@ async def sse_reader(
                             raise sse_exc
                         except Exception as exc:
                             logger.exception("Error in sse_reader")
-                            await read_stream_writer.send(exc)
+                            try:
+                                await read_stream_writer.send(exc)
+                            except anyio.ClosedResourceError:
+                                # Stream already closed during shutdown, which is expected
+                                logger.debug("Stream closed during error handling - ignoring")
                         finally:
                             await read_stream_writer.aclose()
 
@@ -142,11 +146,12 @@ async def post_writer(endpoint_url: str):
                     try:
                         yield read_stream, write_stream
                     finally:
-                        # FIX: Removed manual cancel - anyio task group handles cleanup automatically
-                        # The manual cancel caused: "RuntimeError: Attempted to exit cancel scope
-                        # in a different task than it was entered in"
-                        # When the async context manager exits, the task group's __aexit__
-                        # will properly cancel all child tasks.
+                        # FIX: Removed manual tg.cancel_scope.cancel() that violated anyio lifecycle
+                        # The original code called tg.cancel_scope.cancel() here, but this caused:
+                        # "RuntimeError: Attempted to exit cancel scope in a different task than it was entered in"
+                        #
+                        # The task group's __aexit__ will properly cancel all child tasks when
+                        # this context manager exits, so no manual cancellation is needed.
                         pass
         finally:
             await read_stream_writer.aclose()