Intercom MCP

Support conversations are where a product's real problems are written down, and they are usually the least accessible data a team owns. Intercom's official MCP server opens that archive to an agent over a hosted endpoint, with the workspace's own permissions still applied.

The server exposes 13 tools. The one that shapes how you use it is a universal search tool built around a query DSL rather than a per-object endpoint: you express what you are looking for — conversations matching a topic, in a date range, with a given tag or state — and get back matching conversations and contacts. That design makes multi-step investigation practical ("find every conversation about failed checkout in the last month, then group them by plan tier") without the agent stitching together a dozen narrow calls.

Typical uses that follow from that: summarising recurring complaints before a planning cycle, pulling every ticket that mentions a feature you are about to change, drafting a reply with the customer's full history in context, or auditing how a specific bug was reported across accounts.

Connection: remote HTTP at https://mcp.intercom.com/mcp. There is nothing to install locally — any MCP client that supports remote servers can point at it, and npx mcp-remote bridges clients that only speak stdio.

Authentication: OAuth is the recommended path and the default. Connecting triggers a browser sign-in and no API key setup is needed. A bearer token from your Intercom workspace works as the alternative for headless environments.

Important constraint: the MCP server is currently supported only for US-hosted Intercom workspaces. Teams on EU or Australian data residency cannot use it yet — check your workspace region before planning around it.

Documentation lives on the Intercom developer site and the server is tracked publicly at intercom/intercom-mcp-server.