labs/module4/notebooks/1_mcp.ipynb: AttributeError in call_tool — stale MCP SDK return shape

[ManojRamani]

Summary

Running the call_tool cell in labs/module4/notebooks/1_mcp.ipynb raises:

AttributeError: 'list' object has no attribute 'model_dump_json'

The notebook is built against an older version of the mcp Python SDK where FastMCP.call_tool returned Sequence[ContentBlock]. Current SDK (mcp==1.26.0) returns a tuple (list[ContentBlock], dict) — (content, structured_result). Iterating the tuple yields the list (no .model_dump_json) and the dict, both of which fail.

Repro

1. Branch: main
2. Open labs/module4/notebooks/1_mcp.ipynb
3. Run the cell containing:

   async def call_tool(name, arguments):
       results = await mcp_server.call_tool(name, arguments)
       for result in results:
           print(result.model_dump_json(indent=2))
   asyncio.run(call_tool("get_alerts", {"state": "CA"}))

Environment

• Python 3.12.8
• mcp==1.26.0
• Windows 11 / WSL-less (native)

Root cause

mcp.server.fastmcp.FastMCP.call_tool signature:

async def call_tool(self, name, arguments) -> Sequence[ContentBlock] | dict[str, Any]

But the runtime returns a 2-tuple (content, structured):

type(results): tuple
len(results): 2
  [0] type=list, value=[TextContent(...)]
  [1] type=dict, value={'result': '...'}

Proposed fix

Update the cell to unpack:

async def call_tool(name: str, arguments: Dict[str, Any]) -> None:
    content, structured = await mcp_server.call_tool(name, arguments)
    for result in content:
        print(result.model_dump_json(indent=2))

Also check other call_tool usages in the notebook (and any sibling notebooks) for the same drift.

Scope

Limited to labs/module4/notebooks/1_mcp.ipynb — will expand if the same pattern is present elsewhere after audit.

[ManojRamani]

Update: second defect in the same notebook (Windows + Jupyter)

While fixing the call_tool tuple-unpack issue, verified two additional independent failures in 1_mcp.ipynb cells 13 and 15-16 on Windows + Jupyter:

UnsupportedOperation: fileno

Root cause

mcp.client.stdio.stdio_client defaults errlog=sys.stderr. Under Jupyter/ipykernel, sys.stderr is an OutStream wrapper whose .fileno() raises io.UnsupportedOperation. The MCP SDK's Windows subprocess fallback (_create_windows_fallback_process) then calls subprocess.Popen(..., stderr=<OutStream>) which in turn calls msvcrt.get_osfhandle(stderr.fileno()) → crash.

Affects:

• Cell 13 — direct stdio_client(server_params) call.
• Cell 15-16 — pydantic_ai.mcp.MCPServerStdio which internally calls stdio_client(server=server) with no errlog kwarg.

Scope

I'll bundle all three cell fixes into a single PR against this issue since they're all the same "notebook drifted relative to current SDKs on Windows" class of problem and touch only labs/module4/notebooks/1_mcp.ipynb.

Proposed fixes

• Cell 8 — unpack the tuple from FastMCP.call_tool (as originally described).
• Cell 13 — pass explicit errlog=open(os.devnull, "w") to stdio_client.
• Cell 15 — monkey-patch pydantic_ai.mcp.stdio_client so its internal call gets a safe errlog.

All three verified working locally.