StreamableHTTP: a client disconnect while reading the POST body is logged as an exception and answered 500

Initial Checks

• I confirm that I'm using the newest release of my line (the latest 2.x, or the latest 1.x if I'm still on v1)
• I confirm that I searched for my issue in https://github.com/modelcontextprotocol/python-sdk/issues before opening this issue

Release line

2.x (current stable), and the same on 1.28.1

Description

StreamableHTTPServerTransport._handle_post_request reads the request body inside its blanket except Exception. When the client goes away before the body arrives, await request.body() raises starlette.requests.ClientDisconnect, the transport logs it with logger.exception("Error handling POST request") and answers 500. That is a normal network event on a public MCP endpoint (mobile clients, proxies with short timeouts, health probes), and it lands as an unhandled server error with a stack trace in every log aggregator. #2064 / #2257 handled the disconnect in _handle_message; this is the earlier point, in streamable_http.py (the body = await request.body() line and the except Exception as err below it). The SSE transport has no handling for it either.

Example Code

import asyncio
import importlib.metadata as md
import logging

from mcp.server.streamable_http import StreamableHTTPServerTransport

records = []


class Capture(logging.Handler):
    def emit(self, record):
        records.append(record)


logging.getLogger("mcp").addHandler(Capture())
logging.getLogger("mcp").setLevel(logging.DEBUG)


async def main():
    transport = StreamableHTTPServerTransport(mcp_session_id=None)
    scope = {
        "type": "http", "method": "POST", "path": "/mcp", "query_string": b"", "headers": [
            (b"content-type", b"application/json"),
            (b"accept", b"application/json, text/event-stream"),
            (b"content-length", b"58"),
        ],
    }

    async def receive():  # the client went away before sending its body
        return {"type": "http.disconnect"}

    sent = []

    async def send(message):
        sent.append(message)

    async with transport.connect():
        try:
            await asyncio.wait_for(transport.handle_request(scope, receive, send), timeout=5)
        except asyncio.TimeoutError:
            print("handle_request did not return within 5s")
    status = next((m["status"] for m in sent if m["type"] == "http.response.start"), None)
    print(f"mcp {md.version('mcp')}: response status = {status}")
    for r in records:
        if r.levelno >= logging.ERROR:
            print(f"  logged {r.levelname} by {r.name}: {r.getMessage()} | exc: {type(r.exc_info[1]).__name__ if r.exc_info else None}")


try:
    asyncio.run(asyncio.wait_for(main(), timeout=15))
except asyncio.TimeoutError:
    print("main did not finish within 15s")

Output on 1.28.1 and 2.1.1:

mcp 2.1.1: response status = 500
  logged ERROR by mcp.server.streamable_http: Error handling POST request | exc: ClientDisconnect

(The script caps handle_request at 5 s because the stub send / receive is not a real ASGI server; the 500 and the log line are sent before that.)

Expected: ClientDisconnect caught before the generic handler, logged at debug or info, and no response attempted, since there is nobody to answer.

Python & MCP Python SDK

Python 3.12.13, mcp 1.28.1 and 2.1.1, starlette 1.3.1, macOS

Inizia in mcp/server/streamable_http.py, in StreamableHTTPServerTransport._handle_post_request, prestando particolare attenzione alla riga body = await request.body() e al successivo gestore generico delle eccezioni. Confronta la gestione esistente delle disconnessioni in _handle_message, a cui si fa riferimento in #2064 e #2257. Il lavoro è completato quando ClientDisconnect viene gestito senza uno stack trace di errore né una risposta 500 e la copertura di regressione verifica questo comportamento.