Question

You are building a FastMCP server that exposes tools to an LLM agent for network automation. One tool, apply_interface_description, updates a device interface description via RESTCONF PATCH. During testing you notice the agent sometimes re-invokes this tool multiple times in a single planning loop because it treats every tool as potentially side-effecting and 'checks its work' by calling again, causing unnecessary device writes. Which action best communicates the tool's behavior to the agent to reduce redundant invocations while keeping the safeguard accurate?

Accepted Answer

Annotate the tool with metadata indicating it is idempotent (repeated calls with the same arguments produce the same result), so the agent can reason that re-invocation is unnecessary.

Answer

Set the FastMCP tool's timeout to a very low value so repeated calls fail fast and the agent stops retrying.

Answer

Rename the tool to 'read_interface_description' so the agent stops classifying it as a write operation.

Answer

Lower the LLM temperature to 0 so the agent produces fully deterministic plans and never repeats a tool call.