Question

You are building a FastMCP server that exposes a tool called collect_interface_errors, which SSHes into 40 core switches sequentially and parses error counters. During testing with an AI agent, the tool consistently fails because the agent's LLM client aborts the request after 30 seconds, well before the sequential collection completes (~90 seconds). You must keep the collection accurate but make the tool usable by the agent. Which change BEST addresses this?

Accepted Answer

Refactor the tool to collect the switches concurrently (e.g., asyncio with bounded parallelism) so total execution finishes within the client timeout window, returning the full structured result.

Answer

Increase the SSH connection timeout on each Netmiko session so individual device connections do not drop during the collection.

Answer

Have the tool return a partial result after 25 seconds and instruct the agent to call the tool repeatedly until all switches are covered.

Answer

Disable TLS verification between the agent's LLM client and the MCP server to reduce handshake overhead and speed up the response.