ACE MCP
Answer reporting questions beyond the product’s existing views.
Which calls mention cancellation?
Returns matching call records
Returns counts by tracker
Matching calls + count by tracker
Problem
Customers of ACE, RingCentral’s conversation intelligence product, wanted custom analysis of conversation data without requiring a separate engineering cycle for every report. A Model Context Protocol (MCP) interface offered read-only access from an AI client while respecting the user’s permissions.
My contribution
I designed the query language and client-facing interface around document search and aggregation, and contributed multi-turn evaluations. Another engineer implemented the interface.
Decision
The language needed to express varied reporting requests through a small tool surface. Evaluation had to check sequences of calls, because a later query could depend on an earlier result. Correct individual calls were only part of getting the complete report right.
Outcome
ACE MCP has released. The shared reporting interface covers 23 conversation-intelligence entities.