FastMCP + Pydantic + Future Annotations Issue Reproduction

This is a minimal reproduction of a bug where FastMCP fails to properly resolve custom types when from __future__ import annotations is enabled.

Issue Description

When using from __future__ import annotations, FastMCP's TypeAdapter creation fails to resolve custom types because it uses FastMCP's module globals instead of the function's module globals for type evaluation.

Error

NameError: name 'CustomType' is not defined

Versions Tested

• FastMCP: 2.10.2
• Pydantic: 2.11.7
• Python: 3.11.10

Root Cause

1. from __future__ import annotations converts type annotations to strings
2. FastMCP calls TypeAdapter(method) which triggers Pydantic's type evaluation
3. Pydantic calls eval('CustomType', fastmcp_globals, localns)
4. CustomType is not available in FastMCP's globals, only in the original module's globals
5. Results in NameError

Files

• broken_example.py - Demonstrates the issue
• working_example.py - Shows workarounds
• test_reproduction.py - Test to reproduce the issue

How to Run

pip install fastmcp==2.10.2
python test_reproduction.py

Workarounds

1. Remove from __future__ import annotations
2. Use built-in types like dict[str, Any] instead of custom types
3. Remove return type annotations entirely for FastMCP methods