FastMCP (mcp 1.x) sync tool calling asyncio.run(): 'asyncio.run() cannot be called from a running event loop' + 'coroutine ... was never awaited'; just make the tool async def
Symptom: an MCP server built with from mcp.server.fastmcp import FastMCP (mcp 1.30.0) has a plain def tool that fans out HTTP range requests by defining an inner async def fetch_all() and calling asyncio.run(fetch_all()). When a client calls the tool, the server logs RuntimeWarning: coroutine 'get_flight_summary.<locals>.fetch_all' was never awaited, and the tool result is {"error": "asyncio.run() cannot be called from a running event loop"} (the exception was caught and returned).
Cause: FastMCP runs on an asyncio event loop and calls synchronous tool functions directly on that loop's thread. Inside the tool there is already a running loop, so asyncio.run() raises RuntimeError before it ever awaits the coroutine, which is why the 'never awaited' warning shows up too. This is easy to reproduce outside any client:
mcp = FastMCP("t")
@mcp.tool()
def sync_tool() -> str:
async def inner(): return 1
return str(asyncio.run(inner())) # RuntimeError inside the server
await mcp.call_tool("sync_tool", {})A common conclusion is that 'FastMCP v1 doesn't support async tools'. That's wrong: @mcp.tool() on an async def works in mcp 1.30.0, and the framework awaits it. So the simplest fix is:
@mcp.tool()
async def get_flight_summary(model_id: str, sample_every: int = 10) -> str:
results = await asyncio.gather(*(fetch(url) for url in urls)) # e.g. httpx.AsyncClient
...If the HTTP code is synchronous (requests), the alternative that also works is a concurrent.futures.ThreadPoolExecutor(max_workers=N) with as_completed() inside a plain def tool. Never nest asyncio.run() in a tool.
Note that blocking calls (requests.get, time.sleep) inside a sync tool also block the server's loop while they run. Verify with await mcp.call_tool(...) in a quick script: the async version returns the gathered results with no RuntimeWarning.