How do you find out why an AI agent is calling your tool?

THE SHORT ANSWER

Ask it, with one optional parameter on every tool. Amplitude's MCP server has a Tool Rationale field, a brief explanation of why you are calling this tool, and per Chanaka Perera's post of October 6, 2026 agents fill it in about three quarters of the time without being told to. Pair it with a funnel across tool calls. Of 18,703 Amplitude users who queried data, 20% rendered a chart and less than 5% saved it, and the rationales on the stalled calls showed agents were trying to save. Treat each rationale as a claim and check it against what the agent did next. In May 2026 I wrote that agents don't answer interviews and listed six discovery methods starting with telemetry. This is the seventh, and it is the one that supplies the why.

Ask it, in the tool call itself.

Amplitude's MCP server carries one optional parameter on every tool, called Tool Rationale: a brief explanation of why you are calling this tool. Chanaka Perera's post on consolidating the server reports that agents fill it in about three quarters of the time, unprompted.

The field earned its place on one funnel. Creating a chart took three chained calls. Of 18,703 users who queried data, 20% went on to render a chart and less than 5% saved it. Sequence data alone could not say whether agents wanted only the data or wanted the chart and failed to get there. The rationales on the stalled calls read like "save chart edit for dashboard analysis." Agents were trying to save. Amplitude collapsed the chain into one call.

This corrects something I published. Customer Discovery When Your Customer Is an Agent, from May 2026, says agents don't answer interviews and offers six methods, the first being telemetry and the second being interviews with the human who deployed the agent. Telemetry gives the what. The rationale gives the stated why, at the moment of use, for every customer.

The steps: add the optional parameter, build the funnel across calls, read the rationales where sessions stop, and check each one against what the agent did next, because an agent's account of itself is a claim and not a log.

The full piece is Agents Don't Answer Interviews. Amplitude's Left a Note.

SOURCES

THE LONG VERSION

RELATED ANSWERS

Last reviewed 2026-10-06 · 2 min read