fix: add timeout to MCP tool future.result() to prevent indefinite bl…

[walker0643]

Bug: MCPManager tool calls block indefinitely when MCP server hangs

Summary

When an MCP server becomes unresponsive during a tool call, the ToolClass.call() method in mcp_manager.py blocks the calling thread indefinitely. This causes worker threads to remain permanently active, consuming CPU resources until the process is manually killed.

Root Cause

In create_tool_class(), the result of the async tool call future is awaited with no timeout:

future = asyncio.run_coroutine_threadsafe(client.execute_function(tool_name, tool_args), manager.loop)
try:
    result = future.result()  # blocks forever if the coroutine never resolves
    return result

If the underlying MCP server stalls (slow HTTP request, hung subprocess, failed SSE connection, DNS/SSL issue, etc.) the coroutine in execute_function never resolves, and future.result() waits indefinitely. The thread pool worker that called the tool is permanently blocked, and because it holds a reference to the event loop, it also prevents clean shutdown.

Observed Symptoms

• concurrent.futures thread pool workers stuck in _worker and marked active in process monitors indefinitely
• Python process pegged at 100% CPU on one or more threads for hours with no forward progress
• Process does not respond to normal shutdown; requires SIGKILL
• py-spy dump shows all active threads blocked inside future.result() with no further stack progression

Fix

Add a timeout parameter to future.result() and cancel the future on timeout