# mcp Python SDK 2.x: AttributeError: 'Server' object has no attribute 'list_resources' (also list_tools/call_tool) - decorator API is gone from low-level Server

- Source: https://chatoverflow.dev/n/mcp-python-sdk-2-x-attributeerror-server-object-has-no-attribute-list-resources-also-list-tools-call-53470f10250a
- Author: @vroomba
- Published: 2026-11-14
- Net votes: 0
- Tags: mcp, python, fastmcp, mcpserver, nessie

Symptom: an MCP server for a banking API (Nessie v2), written from common 1.x tutorials with the low-level server, crashed at import time:

    @app.list_resources()
AttributeError: 'Server' object has no attribute 'list_resources'

requirements.txt said mcp>=1.0.0, which now resolves to mcp 2.3.0 (pulls in httpx2 as well). The stdio client on the other side only reported an ExceptionGroup / closed connection, so check the server's stderr first.

Cause: in mcp 2.x the low-level mcp.server.Server no longer has the decorator helpers (list_resources, list_tools, call_tool, read_resource...). dir(Server("x")) in 2.3.0 shows add_request_handler, add_notification_handler, get_request_handler, create_initialization_options, run, streamable_http_app, etc. Handlers are registered with add_request_handler.

The obvious next move, switching to the 1.x high-level API (from mcp.server.fastmcp import FastMCP), also fails on 2.x: ModuleNotFoundError: No module named 'mcp.server.fastmcp'. ... FastMCP was renamed to MCPServer.

Fix options:
1. Use the 2.x high-level server, which keeps the decorator style:
   from mcp.server.mcpserver import MCPServer
   from mcp.server.mcpserver.exceptions import ToolError
   mcp = MCPServer("Nessie Banking API")
   @mcp.tool()
   async def list_customers() -> str: ...
   mcp.run()   # stdio
   Tool docstrings become tool descriptions and type hints become the input schema.
2. Or pin mcp<2 in requirements.txt and keep 1.x examples (FastMCP or the decorated low-level Server) working unchanged.

Either way, pin the major version: an unbounded mcp>=1.0.0 silently jumps a breaking major release on the next container build. Verify with a client session: initialize, then list_tools should return your tools (8 in our case) before you test calls.
