Streamable HTTP transport fails when accessing MCP server on Cloud Run using sdk

[sidgangs99]

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

Description

When using the python-sdk with transport set to streamable-http, the connection to a remote MCP server deployed on Google Cloud Run does not establish. The API stays in a pending state, and no updates are received.

However, the same MCP server works fine under the following conditions:

• When accessed via Google Cloud Proxy
• When transport is changed to SSE
• When using the JS SDK, streamable-http transport works without issues

This suggests a compatibility or configuration issue specifically between python-sdk's streamable-http transport and direct access to Cloud Run URLs.

---

Reproduction Steps

1. Deploy an MCP server to Cloud Run using this official guide:
   https://cloud.google.com/blog/topics/developers-practitioners/build-and-deploy-a-remote-mcp-server-to-google-cloud-run-in-under-10-minutes
2. Use:
   • Transport: streamable-http
   • Endpoint: Public URL of Cloud Run MCP server
3. Attempt to connect using <cloud_run_url>/mcp
4. Observe: API status remains in pending and no data is received

---

Observations

• Switching transport to SSE resolves the issue.
• Using a Cloud Proxy to access the MCP server also works with streamable-http.
• JS SDK does not face this issue under the same deployment conditions.
• The Cloud Run URL is publicly accessible and works with other tools.

---

Environment

• SDK: python-sdk (v1.9.4)
• MCP Server: Deployed on Cloud Run
• Python version: [3.12.11]
• OS - Docker image: [linux/amd64]

---

Expected Behavior

The SDK should establish a connection to the Cloud Run-hosted MCP server and transition the API status from pending to active.

---

Additional Notes

This might point to either:

• SDK's HTTP streaming implementation not being fully compatible with Cloud Run’s networking behavior
• Some internal headers or keep-alive expectations that differ between SSE and streamable-http

Any guidance, workaround, or fix would be appreciated.

Example Code

from mcp.server.fastmcp import FastMCP

# Create server
mcp = FastMCP("Echo Server")


@mcp.tool()
def echo(text: str) -> str:
    """Echo the input text"""
    return text


def main():
    mcp.run(transport="streamable-http", host="0.0.0.0", port=8080)


if __name__ == "__main__":
    main()

Python & MCP Python SDK

SDK Version: 1.9.4
Python Version: 3.12.11

[eiseleMichael]

A quick workaround would be to use FastAPI to wrap your FastMCP server, this is how we are running them currently in Cloud Run.

#main.py

mcp = FastMCP(
    "Demo",
    stateless_http=True,
)

@asynccontextmanager
async def lifespan(_: FastAPI):
    async with contextlib.AsyncExitStack() as stack:
        await stack.enter_async_context(mcp.session_manager.run())
        yield

app = FastAPI(lifespan=lifespan)

and use uvicorn (or any other server) to run it.
In your docker file:

... 

CMD ["uvicorn", "src.main:app", "--port", "8080", "--host", "0.0.0.0"]

[santosarns]

Any update on possible workarounds tried? Still fail to connect trying to wrap FastMCP using fastAPI, when deployed locally, connection has no issues. @sidgangs99

Have you tried this out? @eiseleMichael

[sachin-philip]

@santosarns , @sidgangs99 did you manage to solve this without FastAPI wrapper?