Skip to main content
Agent MCP lets you connect to a single deployed agent from your own tools — ask it questions, retrieve the cited resources behind its answers, and trigger sync/deploy. It works with any MCP-compatible client: Claude, Cursor, agent frameworks, or a custom client of your own. For access to a whole room — pipelines, memories, lenses, and data — use Room MCP instead.

​
Get connected

Open your agent’s Integration & Sharing → Integration tab and copy its MCP URL and Access Token. See Agents. Configure your MCP client with that URL and send the token as a Bearer header:
Treat access tokens like passwords. Don’t commit them to source control — load them from environment variables or a secrets manager.
The room’s MCP Playground is the fastest way to explore before wiring a client up: it lists every tool with its input schema and lets you call them live, showing the exact requests and raw responses an MCP client would see.

​
Available tools

Your client’s tool listing always reflects the live set — descriptions and input schemas come straight from the agent.

​
A typical exchange

1

Ask

Call query with your question and with_citation: true. The agent answers from its configured knowledge, citing the rows and documents it used.
2

Follow the citations

Pass the cited resource IDs to get_resource_download_url to fetch the underlying documents.
3

Keep it fresh

After new data lands, call sync_agent_data_app and then deploy_agent so the agent answers from the latest state.

MCP API Overview

Agent vs. room access and credentials.

Room MCP

The full room surface: pipelines, memories, lenses, agents, and data access.

Agents

Configure and deploy the agent behind this endpoint.

Sharing

Share chats and artifacts without an integration.