FastMCP server with SSE transport fails to shut down on a signal

[growler]

A very simple server fails to shut down on a signal if it processed at least one request:

❯ python test-server.py --port=8085
INFO:     Started server process [216035]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8085 (Press CTRL+C to quit)
INFO:     127.0.0.1:55188 - "GET /sse HTTP/1.1" 200 OK
INFO:     127.0.0.1:55192 - "POST /messages/?session_id=e690de9733914d09aebf2b0e8c78191c HTTP/1.1" 202 Accepted
INFO:     127.0.0.1:55192 - "POST /messages/?session_id=e690de9733914d09aebf2b0e8c78191c HTTP/1.1" 202 Accepted
INFO:     127.0.0.1:55192 - "POST /messages/?session_id=e690de9733914d09aebf2b0e8c78191c HTTP/1.1" 202 Accepted
Processing request of type CallToolRequest
^CINFO:     Shutting down
INFO:     Waiting for background tasks to complete. (CTRL+C to force quit)

The expected behaviour would be:

❯ python test-server.py --port=8085
INFO:     Started server process [216006]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://0.0.0.0:8085 (Press CTRL+C to quit)
^CINFO:     Shutting down
INFO:     Waiting for application shutdown.
INFO:     Application shutdown complete.
INFO:     Finished server process [216006]

Here is the code for both the server and the client.

What do I miss?

the server

import click
import sys
from pydantic import Field
from mcp.server.fastmcp import FastMCP

@click.command()
@click.option("--port", default=8085, help="Port to listen", type=int)
def main(port: int):
    mcp = FastMCP(
        "mcp-echo-tool",
        debug=True,
        log_level="INFO",
        port=port,
    )

    @mcp.tool(name="echo")
    async def echo_tool(string: str = Field(description="A string to echo back")) -> str:
        """Echoes back the input string"""
        return string

    mcp.run(transport='sse')
    return 0

if __name__ == "__main__":
    sys.exit(main())

and the client

import asyncio
import click
import sys
from mcp.client.session import ClientSession
from mcp.client.sse import sse_client

async def client(url, string):
    async with sse_client(url) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            result = await session.call_tool("echo", {"string": string})
            print(result.content)

@click.command()
@click.option('--url', default='http://localhost:8085', help='MCP Server URL', type=str)
@click.argument('string', type=str)
def main(url: str, string: str):
    asyncio.run(client(url, string))

if __name__ == "__main__":
    sys.exit(main())

[withwind8]

I think uvicorn is waiting for close opened sse connection. But I've no idea how to do it with fastmcp.

Can anyone help?

[nadeemcite]

It’s unclear if there’s a manual way to disconnect from the session. The session gets created when you attempt to connect using an MCP client, and even the MCP Inspector doesn’t terminate the session when "Disconnect" is clicked.

[evalstate]

deleted previous message as thought it was raised on my project! - i have been scratching my head on this one...! would be good to get fixed.

[growler]

... I would also like to add that if the MCP server has at least one real established connection, it fails to shut down even with forced Ctrl-C. The only way is to kill the process.

[sealpp]

I'm experiencing the same issue. Initially, I suspected it might be caused by the Auto Forward Ports feature of the VS Code Remote plugin. However, after disabling automatic port forwarding, the problem persists.

[420516460]

so this issue has been resolved?

[PWZER]

Fixed in #586

[pranftw]

Also an automatic timeout would be a useful feature. If the client has connected, but is no longer actively sending messages, then terminate the connection after specified time.

[lanstonchu]

for @pranftw 's suggestion of adding automatic timeout, I think we should add timeout_graceful_shutdown=3 to run_streamable_http_async()?

So it will become:-

config = uvicorn.Config(
    starlette_app,
    host=self.settings.host,
    port=self.settings.port,
    log_level=self.settings.log_level.lower(),
    timeout_graceful_shutdown=3 , # Force shutdown after 3 seconds
)

• ref.: uvicorn settings: https://www.uvicorn.org/settings/#timeouts
  • "--timeout-graceful-shutdown - Maximum number of seconds to wait for graceful shutdown. After this timeout, the server will start terminating requests."

[fourtyplustwo]

Still experiencing this issue with:

Name: mcp Version: 1.14.0 Summary: Model Context Protocol SDK

Cursor is holding the connection and not letting go.. Repeated attempts to Ctrl+C does not help.

[Shanek2k25k]

for @pranftw 's suggestion of adding automatic timeout, I think we should add timeout_graceful_shutdown=3 to run_streamable_http_async()?

So it will become:-

config = uvicorn.Config(
    starlette_app,
    host=self.settings.host,
    port=self.settings.port,
    log_level=self.settings.log_level.lower(),
    timeout_graceful_shutdown=3 , # Force shutdown after 3 seconds
)

• ref.: uvicorn settings: https://www.uvicorn.org/settings/#timeouts

  • "--timeout-graceful-shutdown - Maximum number of seconds to wait for graceful shutdown. After this timeout, the server will start terminating requests."

@lanstonchu Thanks!!!