Data MCP vs Builder MCP
Both are PolyAI-hosted servers you connect in to. They cover different jobs and authenticate differently.To connect a client out from a live agent to a third-party server instead, see Agent Studio MCP integrations.
What you can do
Conversation search
Find conversations by their metrics and scores — filter and sort on quality scores, duration, and outcomes over a date range.
Transcripts
Read the full, turn-by-turn record of a conversation, or search across many conversations for a word or phrase with surrounding context.
Metrics catalog
List the metrics available for an account or project — the built-in ones plus any custom metrics your projects define.
Analytics
Roll a metric up over time into buckets you can chart — grouped, filtered, and thresholded however you like.
Traces and function logs
Open a conversation’s spans and timeline, break latency and errors down per turn, and read the log lines your GenAI functions wrote.
Get started
1. Get an account API key
Every request carries an API key in theX-API-KEY header. Data MCP uses an account API key, created from your account page in Agent Studio — it’s scoped to the account and can query any project in it. See Authentication for details.
2. Set your key and endpoint
Keys are region-specific — a US key doesn’t work against the UK endpoint. Match the region to where your account lives. Self-serve (Studio) accounts are supported too.3. Add the server to your client
- Claude Code
- Cursor
- Claude Desktop
- Codex
claude mcp list, then ask Claude to list its PolyAI data tools.Capabilities
Data MCP exposes 13 tools — conversation search, transcript search, single-conversation fetch, a metrics catalog, metric aggregation, conversation traces and timelines, span search and aggregation, and function log search and aggregation. See Capabilities for each tool and its parameters.Browse the tools
Every Data MCP tool, grouped by what it does, with its parameters.

