Initial Checks

• I confirm that I'm using the latest version of MCP Python SDK
• I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue

Description

When a client cancels a long-running tool call (e.g., user hits Escape in Claude Code), the MCP server crashes with an assertion error and becomes unresponsive.

What I'm seeing:

• Server crashes with "AssertionError: Request already responded to"
• All subsequent requests fail
• Server must be restarted

What I expect:

• Server should handle cancellation gracefully
• Server should remain responsive for subsequent requests

Root cause:
In Server._handle_request, when a CancelledError occurs:

1. RequestResponder.cancel() has already sent an error response
2. The CancelledError propagates to _handle_request
3. It tries to send another error response
4. This triggers "already responded to" assertion

Example Code

from mcp.server import Server
import asyncio

server = Server("test-server")

@server.call_tool()
async def slow_tool(name: str, arguments: dict):
    # Any long-running operation
    await asyncio.sleep(30)
    return [{"type": "text", "text": "Done"}]

# When client calls slow_tool and cancels (e.g., hits Escape in Claude Code)
# Server crashes with: AssertionError: Request already responded to

Python & MCP Python SDK

Python 3.8+ (tested on 3.11)
MCP Python SDK: Latest version from main branch

Related issues: jlowin/fastmcp#508, jlowin/fastmcp#823