Skip to main content
The Model Context Protocol (MCP) is an open standard for connecting AI systems to tools. PolyAI works with MCP in two directions — and it’s worth being clear which one you want before you start, because they solve opposite problems.

Two directions

PolyAI's MCP servers

Connect a client in. Point an MCP client — Cursor, Claude Code, Claude Desktop, Codex — at one of PolyAI’s authenticated, hosted MCP servers to build agents or query their data from your IDE. Your tool acts as a client of PolyAI’s server.

Agent Studio MCP integrations

Consume external tools. Add a third-party MCP server inside Agent Studio so your live agent can call its tools during conversations. This is your agent acting as a client of someone else’s server.
The rest of this section documents PolyAI’s authenticated MCP servers — the ones you connect a client to. For the connect-out direction, see Agent Studio MCP integrations.

PolyAI’s MCP servers

PolyAI hosts two authenticated MCP servers. They cover different jobs, but connect the same way — both use the streamable HTTP transport, run in the same regions, and take the same account API key in the X-API-KEY header.

Builder MCP

Build, test, and deploy PolyAI agents — agents, data, and alerts as tools.

Data MCP

Query conversation data — search conversations, read transcripts, aggregate metrics.

Get connected

1

Authenticate

Create the API key your server needs and configure your client’s auth. See Authentication.
2

Connect a client

Add the server to Cursor, Claude Code, Claude Desktop, or Codex with a single command. See Quickstart.
3

Build or query

Use the Builder MCP to create, test, and deploy agents, or the Data MCP to query conversation data — all conversationally.
Last modified on July 17, 2026