TL;DR
Intercom's official MCP exposes 13 tools covering conversations, tickets, contacts, companies, and articles hosted inside Intercom. It pairs with Fin, Intercom's AI agent (which runs on Claude under the hood, plus a proprietary Fin Apex model on top). Brainfish MCP, often evaluated as an Intercom MCP alternative for cross-source knowledge work, exposes the authored knowledge layer across every source you've connected - Confluence, Notion, Google Drive, Guru, Helpjuice, ReadMe, Mintlify, OpenAPI specs, websites - plus your Brainfish workspace itself. If your knowledge lives entirely in Intercom articles, Intercom's MCP is a clean fit. If your knowledge lives across multiple systems and you want one layer Claude can search, draft into, and audit, use Brainfish MCP. As of April 2026, the most common pattern in mature CX stacks is to run both.
At a glance
Category Intercom MCP Brainfish MCP Primary object Conversations, contacts, and Intercom articles Authored knowledge across connected sources Tools exposed 13 tools, including universal search, conversation management, and article CRUD Search, read, create, update, draft, audit, coverage testing, answer generation, and follow-up generation Knowledge sources Intercom-hosted articles only Confluence, Notion, Google Drive, Guru, Helpjuice, ReadMe, Mintlify, OpenAPI, and websites Works with teams not on Intercom No — requires an Intercom workspace Yes — Brainfish sits alongside Zendesk, Intercom, Freshdesk, Salesforce, Slack, and Teams Model strategy Fin runs Claude plus proprietary Fin Apex Your Claude — you own the model relationship Regional availability US-hosted workspaces only (as of April 2026) Global Pricing model Bundled with the Intercom Fin AI suite Tiered platform pricing — see brainfishai.com/pricing Best for Intercom-native teams who want Claude to manage their Intercom stack Teams whose knowledge lives across multiple tools and want Claude to operate on all of it
The model-strategy difference
Intercom's public position is that Fin outperforms frontier models at support resolutions - they launched Fin Apex 1.0 as a proprietary post-trained model that "beats GPT-5.4 and Claude Sonnet 4.6." Fin 2 uses Claude under the hood, but you don't point your Claude at Intercom. Intercom points its Claude at your data.
Brainfish inverts that: you keep the direct Claude relationship, and Brainfish is the knowledge layer your Claude reads from and writes to. For engineering-led CX teams who've standardised on Claude (or Cursor, or VS Code Copilot), that matters - your tooling stays yours.
Choose Brainfish MCP when…
- Your knowledge lives across multiple systems (Confluence for engineering, Notion for product, a dedicated help center for customers) and you want Claude to search and edit all of it.
- You want direct Claude control - the buyer owns the Claude relationship, not the vendor.
- You're already on a non-Intercom help desk (Zendesk, Freshdesk, Gorgias, Kustomer) but want the same "AI inside your KB" workflow.
- You want bidirectional article editing from Claude with draft suggestions, not just basic CRUD.
Choose Intercom MCP when…
- Your customer knowledge is fully hosted inside Intercom articles.
- Your primary Claude workflow is conversation and contact management, not authoring.
- You're already on Fin and want the MCP as an extension of existing Intercom AI workflows.
- You're in a US-hosted Intercom workspace (required as of April 2026).
Use both when…
- You want Claude to manage Intercom conversations and operate on the broader knowledge layer behind them. MCP clients can connect to multiple servers in one session - Brainfish handles cross-source knowledge ops, Intercom handles conversations and Intercom-hosted articles.
Where Brainfish MCP isn't the right fit
- Your entire customer-facing knowledge already lives inside Intercom articles and you don't sync from upstream tools.
- You only need Claude to read and respond to conversations - the Intercom MCP alone is enough.
- You're early-stage and the cost of running a second MCP server isn't justified by your CX volume yet.