ValueError in Dependency.__aexit__ when concurrent Docket background tasks share a default instance that stores a ContextVar token

[carterthayer]

What happened?

Any Dependency subclass that calls ContextVar.set() in __aenter__ and reset() in __aexit__, storing the token on self, raises ValueError when two or more Docket background tasks run concurrently. This affects _CurrentContext and _CurrentAccessToken when auth is configured.

Docket interprets the teardown exception as task failure and requeues tasks that actually completed successfully.

ValueError: <Token var=<ContextVar name='_marker' ...>> was created in a different Context

Example Code

# main.py — run with FASTMCP_DOCKET_URL=redis://localhost:6379/0, FASTMCP_DOCKET_CONCURRENCY=5
import asyncio
from contextvars import ContextVar, Token
from fastmcp import FastMCP
from fastmcp.server.tasks import TaskConfig
from uncalled_for import Dependency

_marker: ContextVar[str] = ContextVar("_marker")

class Marker(Dependency["Marker"]):
    _token: Token[str] | None = None
    async def __aenter__(self) -> "Marker":
        self._token = _marker.set("active")
        return self
    async def __aexit__(self, *args: object) -> None:
        if self._token is not None:
            _marker.reset(self._token)
            self._token = None

mcp = FastMCP("bug-repro")

@mcp.tool(task=TaskConfig(mode="optional"))
async def my_tool(marker: Marker = Marker()) -> str:
    await asyncio.sleep(3)
    return "done"

app = mcp.http_app()


# repro.py — submit 4 concurrent background tasks
import asyncio
from fastmcp import Client
from fastmcp.client.transports import StreamableHttpTransport

async def main():
    async with Client(StreamableHttpTransport("http://localhost:8080/mcp")) as client:
        await asyncio.gather(*[client.call_tool("my_tool", {}, task=True) for _ in range(4)])

asyncio.run(main())

Version Information

FastMCP version:                                                                         3.1.1
MCP version:                                                                            1.26.0
Python version:                                                                         3.14.0
Platform:                                                  macOS-26.3.1-arm64-arm-64bit-Mach-O

Also:
pydocket 0.18.2