Repository navigation
Lifespan enters and exits for each request when stateless_http=True #1304
Description
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 stateless_http=True, the lifespan of the MCP server enters and exits with each request. Surely this is unintended behavior?
Example Code
from mcp.server.fastmcp import Context, FastMCP
from contextlib import asynccontextmanager
@asynccontextmanager
async def tmp_lifespan(server: FastMCP):
print(f"Starting lifespan...")
yield
print(f"Ending lifespan...")
mcp = FastMCP("localmcp", host="127.0.0.1", lifespan=tmp_lifespan, port=8000, stateless_http=True)
# Add an addition tool
@mcp.tool()
async def add(a: int, b: int) -> int:
"""Add two numbers"""
return a + b
if __name__ == "__main__":
print(f"Starting MCP server...")
mcp.run(transport="streamable-http")Python & MCP Python SDK
1.13.0
Activity
felixweinberger commentedon Oct 3, 2025
This is actually by design as stateless mode is currently spinning up a new transport for each request
python-sdk/src/mcp/server/streamable_http_manager.py
Lines 161 to 162 in c260e29
Likely could be improved though, marking this as something for potential enhancements in v2
Kludex commentedon Dec 4, 2025
This is a bug.
jcasse commentedon Dec 5, 2025
@savvaki try the low level API. It doesn't have this problem.
felixweinberger commentedon Jan 15, 2026
Working on this as part of Python SDK v2 hackathon.
maxisbey commentedon Feb 19, 2026
Subsumed by #2113 which redesigns lifespan with proper server-scoped vs session-scoped lifetimes.
feat(server): integrate server lifespan into Starlette app lifespan
docs(migration): add lifespan redesign migration guide for Option B
perf(mcp): cache TickTick client across requests on Vercel (#1)