[BUG] Union types with arrays generate invalid MCP schemas (missing 'items' property)

[combiz]

Describe the bug

When FastAPI endpoints use Pydantic models with Union types containing arrays (e.g., Union[str, List[str]]), the generated MCP tool schema is missing the required items property for the array type. This causes LLMs like OpenAI to reject the schema with: "Invalid schema for function 'xxx': In context=('properties', 'text'), array schema missing items."

To Reproduce

1. Create a FastAPI app with a Union type parameter:

  from fastapi import FastAPI
  from pydantic import BaseModel
  from typing import Union, List
  from fastapi_mcp import FastApiMCP

  app = FastAPI()

  class TextRequest(BaseModel):
      text: Union[str, List[str]]  # Union with array type
      keywords: List[str]

  @app.post("/process")
  def process_text(request: TextRequest):
      return {"processed": True}

  mcp = FastApiMCP(app)

2. Connect an LLM (e.g., via LiteLLM) to the MCP server
3. The LLM will fail to call the tool with error:

  litellm.BadRequestError: OpenAIException - Error code: 400 - {
    'error': {
      'message': "Invalid schema for function 'process_text_process_post':
                  In context=('properties', 'text'), array schema missing items.",
      'type': 'invalid_request_error',
      'param': 'tools[0].function.parameters',
      'code': 'invalid_function_parameters'
    }
  }

System Info

• OS: Ubuntu 22.04
• Python version: 3.12.6
• fastapi-mcp version: 0.4.0
• FastAPI version: 0.115.6
• Pydantic version: 2.10.4
• MCP version: 1.1.3
• LLM Provider: OpenAI (via LiteLLM)