Make FallbackProcess cleanup more robust with try-except

- Keep process termination in __aexit__ but wrap in try-except
- Use non-blocking wait with short timeout (0.5s) to prevent hangs
- Catch ProcessLookupError and OSError for already-dead processes
- Still clean up all streams regardless of termination success

This approach is more conventional for a context manager while preventing
failures when the process is already being handled by stdio_client's
cleanup sequence or has already terminated.

Reported-by: Felix Weinberger

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: felixweinberger

src/mcp/client/stdio/win32.py

@@ -73,11 +73,21 @@ async def __aexit__(
         exc_val: BaseException | None,
         exc_tb: object | None,
     ) -> None:
-        """Clean up streams without terminating the process.
+        """Clean up process and streams.
 
-        The process termination is handled by stdio_client's cleanup sequence
-        which implements the MCP spec-compliant shutdown flow.
+        Attempts to terminate the process, but doesn't fail if termination
+        is not possible (e.g., process already dead or being handled elsewhere).
         """
+        # Try to terminate the process, but don't fail if it's already being handled
+        try:
+            self.popen.terminate()
+            # Use our non-blocking wait with a short timeout
+            with anyio.move_on_after(0.5):
+                await self.wait()
+        except (ProcessLookupError, OSError):
+            # Process already dead or being handled elsewhere
+            pass
+
         # Close the file handles to prevent ResourceWarning
         if self.stdin:
             await self.stdin.aclose()

tests/client/test_stdio.py

@@ -156,8 +156,6 @@ async def test_stdio_client_universal_cleanup():
 
     start_time = time.time()
 
-    # Use move_on_after which is more reliable for cleanup scenarios
-    # Increased timeout to account for Windows process termination overhead
     with anyio.move_on_after(8.0) as cancel_scope:
         async with stdio_client(server_params) as (read_stream, write_stream):
             # Immediately exit - this triggers cleanup while process is still running
@@ -166,7 +164,6 @@ async def test_stdio_client_universal_cleanup():
         end_time = time.time()
         elapsed = end_time - start_time
 
-        # Key assertion: Should complete within reasonable time due to timeout mechanism
         # On Windows: 2s (stdin wait) + 2s (terminate wait) + overhead = ~5s expected
         assert elapsed < 6.0, (
             f"stdio_client cleanup took {elapsed:.1f} seconds, expected < 6.0 seconds. "