MCP SSE server: "Received request before initialization was complete" after redeploy

Question

I serve a FastMCP SSE server (modelcontextprotocol/python-sdk, April 2025) with uvicorn. On first run, the MCP Inspector lists tools fine, but after I reload the server with the same command, the next tools/list or tools/call times out and the server logs 'RuntimeError: Received request before initialization was complete'. What causes this and what are the workarounds?

Originally asked on GitHub: https://github.com/modelcontextprotocol/python-sdk/issues/423

AgentArmorHQ team · Accepted Answer

Still open upstream as of the thread's last comment (May 2025). The error means the server received a JSON-RPC request (`tools/list`, `tools/call`, ...) for a session that never completed the required `initialize` handshake. Commenters traced this to clients/proxies that send `initialize` twice, or post to a session that died when the server was restarted.

What worked in the thread:

1. **Python client: call `initialize()` explicitly.** If you connect with `mcp.client.sse.sse_client` + `ClientSession`, add `await session.initialize()` right after creating the `ClientSession` and before any tool call. One commenter confirmed this reliably fixed the error.

2. **Use `mcp-proxy` instead of supergateway.** Per a commenter, supergateway sends its own `initialize` message on launch and then forwards the client's real one, so the server receives `initialize` twice. The same example worked flawlessly with `mcp-proxy`. A later comment noted the double-initialization bug was patched in supergateway 2.7.0, but another user still saw the error after a server reload even with 2.7.0, because some clients (anything-llm, fastagent) don't re-initialize the session after a disconnect.

3. **Temporary server-side patch.** Monkeypatch `ServerSession._received_request` to swallow the `RuntimeError`:

```python
from mcp.server.fastmcp import FastMCP
from mcp.server.session import ServerSession

old = ServerSession._received_request

async def _received_request(self, *args, **kwargs):
    try:
        return await old(self, *args, **kwargs)
    except RuntimeError:
        pass

ServerSession._received_request = _received_request
```

The author of this patch warned it is not a long-term fix, since it hides other `RuntimeError`s. A follow-up comment confirmed the error no longer appeared with it.

AgentArmorHQ team · Answer

**What @wilson-urdaneta tried on GitHub** (Worked)

Added an explicit `await session.initialize()` right after creating the `ClientSession` in a Python SSE client before calling tools; the error stopped occurring.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/423#issuecomment-2807595031

AgentArmorHQ team · Answer

**What @mheubi tried on GitHub** (Worked)

Switched from supergateway to `mcp-proxy`; the same FastMCP example then worked flawlessly, confirming supergateway was sending a second `initialize` against the protocol spec.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/423#issuecomment-2784989849

AgentArmorHQ team · Answer

**What @folkvir tried on GitHub** (Worked)

Monkeypatched `ServerSession._received_request` to catch and ignore the `RuntimeError`; a follow-up comment confirmed the error no longer appeared.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/423#issuecomment-2799890581

AgentArmorHQ team · Answer

**What @Nedomas tried on GitHub** (Partly worked)

Upgraded to supergateway 2.7.0, which patched the double-initialization bug; a later comment still got the error after reloading the server with that version.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/423#issuecomment-2811193602

AgentArmorHQ team · Answer

**What a GitHub user tried on GitHub** (Didn't work)

Reproduced with the FastMCP BMI/weather example under `uv run uvicorn main:app --port=8001 --host=0.0.0.0 --timeout-graceful-shutdown 5`: the Inspector listed tools on first run, but after reloading the server, listing tools timed out with 'Received request before initialization was complete'.

Source: https://github.com/modelcontextprotocol/python-sdk/issues/423

MCP SSE server: "Received request before initialization was complete" after redeploy

6 answers

Add your answer or comment

MCP SSE server: "Received request before initialization was complete" after redeploy

6 answers

Add your answer or comment

Related questions