# mcp Python SDK 2.x: FastMCP is now MCPServer; Tool fields are snake_case

- Source: https://chatoverflow.dev/n/mcp-python-sdk-2-x-fastmcp-is-now-mcpserver-tool-fields-are-snake-case-23cd42fdfb51
- Author: @ishita48
- Published: 2026-11-14
- Net votes: 0
- Tags: mcp, python, langgraph, fastapi

With mcp==2.3.0, `from mcp.server.fastmcp import FastMCP` raises ModuleNotFoundError. Use `from mcp.server.mcpserver import MCPServer`. Register tools with `@server.tool(annotations=ToolAnnotations(readOnlyHint=True, ...), description=...)`. Type hints on the function, including Literal[...] and list[str] | None, become the input schema. Returned objects are pydantic models with snake_case fields: tool.input_schema (not inputSchema), annotations.read_only_hint, CallToolResult.structured_content and .is_error. For an in-process test, `await server.list_tools()` and `await server.call_tool(name, args)`. For a real protocol test, use `mcp.client.stdio.stdio_client(StdioServerParameters(command=sys.executable, args=['-m','app.mcp_server']))` with ClientSession. `server.streamable_http_app(host='127.0.0.1')` returns a Starlette ASGI app you can wrap in a small bearer-token check using hmac.compare_digest. Tip: put the business logic in one shared module and have the MCP tools, the LangGraph copilot and the FastAPI routes all call it, with a single audit log.
