feat: add request cancellation and cleanup

src/mcp/server/lowlevel/server.py

@@ -453,10 +453,15 @@ async def run(
                     logger.debug(f"Received message: {message}")
 
                     match message:
-                        case RequestResponder(request=types.ClientRequest(root=req)):
-                            await self._handle_request(
-                                message, req, session, raise_exceptions
-                            )
+                        case (
+                            RequestResponder(
+                                request=types.ClientRequest(root=req)
+                            ) as responder
+                        ):
+                            with responder:
+                                await self._handle_request(
+                                    message, req, session, raise_exceptions
+                                )
                         case types.ClientNotification(root=notify):
                             await self._handle_notification(notify)
 

src/mcp/server/session.py

@@ -126,19 +126,20 @@ async def _received_request(
             case types.InitializeRequest(params=params):
                 self._initialization_state = InitializationState.Initializing
                 self._client_params = params
-                await responder.respond(
-                    types.ServerResult(
-                        types.InitializeResult(
-                            protocolVersion=types.LATEST_PROTOCOL_VERSION,
-                            capabilities=self._init_options.capabilities,
-                            serverInfo=types.Implementation(
-                                name=self._init_options.server_name,
-                                version=self._init_options.server_version,
-                            ),
-                            instructions=self._init_options.instructions,
+                with responder:
+                    await responder.respond(
+                        types.ServerResult(
+                            types.InitializeResult(
+                                protocolVersion=types.LATEST_PROTOCOL_VERSION,
+                                capabilities=self._init_options.capabilities,
+                                serverInfo=types.Implementation(
+                                    name=self._init_options.server_name,
+                                    version=self._init_options.server_version,
+                                ),
+                                instructions=self._init_options.instructions,
+                            )
                         )
                     )
-                )
             case _:
                 if self._initialization_state != InitializationState.Initialized:
                     raise RuntimeError(

src/mcp/shared/session.py

@@ -1,6 +1,7 @@
+import logging
 from contextlib import AbstractAsyncContextManager
 from datetime import timedelta
-from typing import Generic, TypeVar
+from typing import Any, Callable, Generic, TypeVar
 
 import anyio
 import anyio.lowlevel
@@ -10,6 +11,7 @@
 
 from mcp.shared.exceptions import McpError
 from mcp.types import (
+    CancelledNotification,
     ClientNotification,
     ClientRequest,
     ClientResult,
@@ -38,27 +40,98 @@
 
 
 class RequestResponder(Generic[ReceiveRequestT, SendResultT]):
+    """Handles responding to MCP requests and manages request lifecycle.
+
+    This class MUST be used as a context manager to ensure proper cleanup and
+    cancellation handling:
+
+    Example:
+        with request_responder as resp:
+            await resp.respond(result)
+
+    The context manager ensures:
+    1. Proper cancellation scope setup and cleanup
+    2. Request completion tracking
+    3. Cleanup of in-flight requests
+    """
+
     def __init__(
         self,
         request_id: RequestId,
         request_meta: RequestParams.Meta | None,
         request: ReceiveRequestT,
         session: "BaseSession",
+        on_complete: Callable[["RequestResponder[ReceiveRequestT, SendResultT]"], Any],
     ) -> None:
         self.request_id = request_id
         self.request_meta = request_meta
         self.request = request
         self._session = session
-        self._responded = False
+        self._completed = False
+        self._cancel_scope = anyio.CancelScope()
+        self._on_complete = on_complete
+        self._entered = False  # Track if we're in a context manager
+
+    def __enter__(self) -> "RequestResponder[ReceiveRequestT, SendResultT]":
+        """Enter the context manager, enabling request cancellation tracking."""
+        self._entered = True
+        self._cancel_scope = anyio.CancelScope()
+        self._cancel_scope.__enter__()
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb) -> None:
+        """Exit the context manager, performing cleanup and notifying completion."""
+        try:
+            if self._completed:
+                self._on_complete(self)
+        finally:
+            self._entered = False
+            if not self._cancel_scope:
+                raise RuntimeError("No active cancel scope")
+            self._cancel_scope.__exit__(exc_type, exc_val, exc_tb)
 
     async def respond(self, response: SendResultT | ErrorData) -> None:
-        assert not self._responded, "Request already responded to"
-        self._responded = True
+        """Send a response for this request.
+
+        Must be called within a context manager block.
+        Raises:
+            RuntimeError: If not used within a context manager
+            AssertionError: If request was already responded to
+        """
+        if not self._entered:
+            raise RuntimeError("RequestResponder must be used as a context manager")
+        assert not self._completed, "Request already responded to"
+
+        if not self.cancelled:
+            self._completed = True
+
+            await self._session._send_response(
+                request_id=self.request_id, response=response
+            )
+
+    async def cancel(self) -> None:
+        """Cancel this request and mark it as completed."""
+        if not self._entered:
+            raise RuntimeError("RequestResponder must be used as a context manager")
+        if not self._cancel_scope:
+            raise RuntimeError("No active cancel scope")
 
+        self._cancel_scope.cancel()
+        self._completed = True  # Mark as completed so it's removed from in_flight
+        # Send an error response to indicate cancellation
         await self._session._send_response(
-            request_id=self.request_id, response=response
+            request_id=self.request_id,
+            response=ErrorData(code=0, message="Request cancelled", data=None),
         )
 
+    @property
+    def in_flight(self) -> bool:
+        return not self._completed and not self.cancelled
+
+    @property
+    def cancelled(self) -> bool:
+        return self._cancel_scope is not None and self._cancel_scope.cancel_called
+
 
 class BaseSession(
     AbstractAsyncContextManager,
@@ -82,6 +155,7 @@ class BaseSession(
         RequestId, MemoryObjectSendStream[JSONRPCResponse | JSONRPCError]
     ]
     _request_id: int
+    _in_flight: dict[RequestId, RequestResponder[ReceiveRequestT, SendResultT]]
 
     def __init__(
         self,
@@ -99,6 +173,7 @@ def __init__(
         self._receive_request_type = receive_request_type
         self._receive_notification_type = receive_notification_type
         self._read_timeout_seconds = read_timeout_seconds
+        self._in_flight = {}
 
         self._incoming_message_stream_writer, self._incoming_message_stream_reader = (
             anyio.create_memory_object_stream[
@@ -219,27 +294,45 @@ async def _receive_loop(self) -> None:
                             by_alias=True, mode="json", exclude_none=True
                         )
                     )
+
                     responder = RequestResponder(
                         request_id=message.root.id,
                         request_meta=validated_request.root.params.meta
                         if validated_request.root.params
                         else None,
                         request=validated_request,
                         session=self,
+                        on_complete=lambda r: self._in_flight.pop(r.request_id, None),
                     )
 
+                    self._in_flight[responder.request_id] = responder
                     await self._received_request(responder)
-                    if not responder._responded:
+                    if not responder._completed:
                         await self._incoming_message_stream_writer.send(responder)
+
                 elif isinstance(message.root, JSONRPCNotification):
-                    notification = self._receive_notification_type.model_validate(
-                        message.root.model_dump(
-                            by_alias=True, mode="json", exclude_none=True
+                    try:
+                        notification = self._receive_notification_type.model_validate(
+                            message.root.model_dump(
+                                by_alias=True, mode="json", exclude_none=True
+                            )
+                        )
+                        # Handle cancellation notifications
+                        if isinstance(notification.root, CancelledNotification):
+                            cancelled_id = notification.root.params.requestId
+                            if cancelled_id in self._in_flight:
+                                await self._in_flight[cancelled_id].cancel()
+                        else:
+                            await self._received_notification(notification)
+                            await self._incoming_message_stream_writer.send(
+                                notification
+                            )
+                    except Exception as e:
+                        # For other validation errors, log and continue
+                        logging.warning(
+                            f"Failed to validate notification: {e}. "
+                            f"Message was: {message.root}"
                         )
-                    )
-
-                    await self._received_notification(notification)
-                    await self._incoming_message_stream_writer.send(notification)
                 else:  # Response or error
                     stream = self._response_streams.pop(message.root.id, None)
                     if stream:

tests/issues/test_88_random_error.py

@@ -0,0 +1,111 @@
+"""Test to reproduce issue #88: Random error thrown on response."""
+
+from datetime import timedelta
+from pathlib import Path
+from typing import Sequence
+
+import anyio
+import pytest
+
+from mcp.client.session import ClientSession
+from mcp.server.lowlevel import Server
+from mcp.shared.exceptions import McpError
+from mcp.types import (
+    EmbeddedResource,
+    ImageContent,
+    TextContent,
+)
+
+
+@pytest.mark.anyio
+async def test_notification_validation_error(tmp_path: Path):
+    """Test that timeouts are handled gracefully and don't break the server.
+
+    This test verifies that when a client request times out:
+    1. The server task stays alive
+    2. The server can still handle new requests
+    3. The client can make new requests
+    4. No resources are leaked
+    """
+
+    server = Server(name="test")
+    request_count = 0
+    slow_request_started = anyio.Event()
+    slow_request_complete = anyio.Event()
+
+    @server.call_tool()
+    async def slow_tool(
+        name: str, arg
+    ) -> Sequence[TextContent | ImageContent | EmbeddedResource]:
+        nonlocal request_count
+        request_count += 1
+
+        if name == "slow":
+            # Signal that slow request has started
+            slow_request_started.set()
+            # Long enough to ensure timeout
+            await anyio.sleep(0.2)
+            # Signal completion
+            slow_request_complete.set()
+            return [TextContent(type="text", text=f"slow {request_count}")]
+        elif name == "fast":
+            # Fast enough to complete before timeout
+            await anyio.sleep(0.01)
+            return [TextContent(type="text", text=f"fast {request_count}")]
+        return [TextContent(type="text", text=f"unknown {request_count}")]
+
+    async def server_handler(read_stream, write_stream):
+        await server.run(
+            read_stream,
+            write_stream,
+            server.create_initialization_options(),
+            raise_exceptions=True,
+        )
+
+    async def client(read_stream, write_stream):
+        # Use a timeout that's:
+        # - Long enough for fast operations (>10ms)
+        # - Short enough for slow operations (<200ms)
+        # - Not too short to avoid flakiness
+        async with ClientSession(
+            read_stream, write_stream, read_timeout_seconds=timedelta(milliseconds=50)
+        ) as session:
+            await session.initialize()
+
+            # First call should work (fast operation)
+            result = await session.call_tool("fast")
+            assert result.content == [TextContent(type="text", text="fast 1")]
+            assert not slow_request_complete.is_set()
+
+            # Second call should timeout (slow operation)
+            with pytest.raises(McpError) as exc_info:
+                await session.call_tool("slow")
+            assert "Timed out while waiting" in str(exc_info.value)
+
+            # Wait for slow request to complete in the background
+            with anyio.fail_after(1):  # Timeout after 1 second
+                await slow_request_complete.wait()
+
+            # Third call should work (fast operation),
+            # proving server is still responsive
+            result = await session.call_tool("fast")
+            assert result.content == [TextContent(type="text", text="fast 3")]
+
+    # Run server and client in separate task groups to avoid cancellation
+    server_writer, server_reader = anyio.create_memory_object_stream(1)
+    client_writer, client_reader = anyio.create_memory_object_stream(1)
+
+    server_ready = anyio.Event()
+
+    async def wrapped_server_handler(read_stream, write_stream):
+        server_ready.set()
+        await server_handler(read_stream, write_stream)
+
+    async with anyio.create_task_group() as tg:
+        tg.start_soon(wrapped_server_handler, server_reader, client_writer)
+        # Wait for server to start and initialize
+        with anyio.fail_after(1):  # Timeout after 1 second
+            await server_ready.wait()
+        # Run client in a separate task to avoid cancellation
+        async with anyio.create_task_group() as client_tg:
+            client_tg.start_soon(client, client_reader, server_writer)