Brainfish + Claude: AI knowledge base integration via MCP

Every time your team opens Claude, they start from scratch. Pasting in articles. Re-briefing the AI. Losing the thread the moment they close a tab. Your knowledge base is the system that holds your company's product knowledge, your terminology, your support playbooks, your edge cases. Until now, Claude couldn't read it, couldn't write to it, and couldn't help you keep it current.

Brainfish + Claude changes that. The Brainfish MCP server connects Claude (and Cursor, and VS Code Copilot) directly to your Brainfish knowledge base. Search it, read from it, draft into it, audit it for gaps, push updates back into it. All from inside the AI assistant your team already has open. No copy-pasting. No starting over. The first MCP server purpose-built for the authored knowledge layer behind AI customer service.

See all Brainfish integrations · Book a demo

Why connect Brainfish with Claude?

Connecting Brainfish to Claude turns your knowledge base from a read-only website into a working surface inside your AI assistant. Your team can search articles, find gaps, audit for stale content, draft new articles from a meeting note, and push updates back to Brainfish without ever switching tabs.

What you get:

• Live, sourced answers from your Brainfish knowledge base inside Claude
• Drafting and updating articles directly from Claude
• Coverage testing: ask Claude where customers would get a poor answer before they do
• Quarterly KB audits run as a single conversation
• OAuth setup in Claude.ai in under 60 seconds; JSON config for Cursor and Claude Desktop

This is what the Brainfish AI Support Agents platform looks like for the people who maintain it. The same knowledge powers your customer-facing answers, your support widget, and now the Claude conversation your team is already in. See Auto-Updating Docs for how the upstream stays clean.

How AI customer service works with Brainfish and Claude

The integration uses the Model Context Protocol (MCP), an open standard for connecting AI assistants to live business systems. Setup follows four steps.

1. Connect. From Brainfish admin settings, generate your Agent Key. The Brainfish MCP server lives at https://mcp.brainfi.sh/mcp. In Claude.ai, authenticate via OAuth in under 60 seconds. For Cursor or Claude Desktop, paste the JSON config Brainfish provides.
2. Index. Brainfish exposes your articles, content suggestions, and audit data as MCP resources and tools. The full content sync from Confluence, Google Drive, Notion, Guru, Helpjuice, and other sources continues to run via Brainfish Content Sync integrations.
3. Use. Ask Claude to search the KB, read a specific article, find duplicates, draft a new article from a meeting note, or coverage-test against a question set. Claude calls the right Brainfish tool and writes results back into your workspace.
4. Review and ship. Drafted articles land in Brainfish as drafts for human review before publish. Audit findings come with reasoning so your team knows why a flag was raised.

The pattern is simple: ask in Claude, fix in Claude, publish from Claude. Read more about why Brainfish builds for the knowledge layer.

What are the benefits of using Brainfish with Claude?

For knowledge managers and CS enablement leads. KB maintenance happens inside Claude instead of an admin UI. Quarterly reviews compress into a single conversation. Drafts move from meeting note to published article without copy-pasting between three tools.

For support and CX leaders. Every downstream answer (in your widget, your help center, your Slack and Teams agents) gets fresher input. The freshness loop closes. See for your support and CX team for how this rolls up to ticket deflection.

For product and engineering teams. Build-vs-buy gets easier. Building this internally means owning a knowledge pipeline and an MCP server on top. With Brainfish you get both, plus the agent layer, plus the analytics. The CTO conversation moves from "should we build this" to "let's stop maintaining what we already have." See for your product team.

For your customers. First answers stay accurate because the knowledge feeding them stays current. See for your users.

How is this different from using Claude alone (or other MCP servers)?

Claude on its own is a strong reasoning model. It just doesn't know your business. Pasted articles fix that for one session and disappear when the tab closes. The next session starts from zero.

Other MCP servers in the support category (Intercom's official MCP, community-built Zendesk MCPs) expose ticket and conversation data. They tell Claude what customers asked. They don't tell Claude what the right answer is, where the gaps are, or how to write a new article that matches your voice.

Brainfish MCP is the first server purpose-built for the authored knowledge layer.

If you've been pasting articles into Claude every morning, that's the entire problem this fixes. See The Chatbot Graveyard for why static knowledge is the underlying issue.

When is the Brainfish + Claude integration most valuable?

This integration earns its keep when your knowledge base is moving faster than your team can manually maintain. The strongest fits:

• Teams shipping product updates weekly or faster, where docs always lag releases
• CS teams running multi-product or multi-brand help centers with constant content drift
• Support orgs already standardized on Claude for triage, drafting, or QA
• Founders, COOs, and CTOs evaluating whether to build a knowledge pipeline in-house
• Heads of CS Enablement responsible for ramping new joiners against a moving target

If your team uses Claude daily and re-briefs it every session, this is the integration that ends that loop.

How do I set up the Brainfish + Claude integration?

1. Get an Agent Key. From Brainfish admin settings, create or copy your Agent Key. The key is required for answer generation and follow-up tools.
2. Open the MCP setup screen. In Brainfish, go to Integrations > Claude / MCP. Copy the MCP endpoint URL: https://mcp.brainfi.sh/mcp
3. Connect Claude. In Claude.ai, paste the endpoint and authenticate via OAuth. Setup runs in under 60 seconds.
4. (Optional) Connect Cursor or Claude Desktop. Drop the JSON config Brainfish provides into the MCP settings of either client.
5. Verify. Ask Claude: "Search my Brainfish KB for refund policy." If you see a sourced answer, you're live.

For the full walkthrough, see the Brainfish MCP setup guide.

What results can I expect from Brainfish + Claude?

Customers running Brainfish MCP today are already changing how they work.

• Buffer uses MCP to draft and update Help Center articles inside Claude. Product updates ship with documentation the same day, not weeks later.
• Hearth is building a real-time CS conversation guide grounded live in Brainfish, so every agent on the team responds like the most experienced one.
• Relevance AI uses a dual-agent architecture with Brainfish as the knowledge backbone, running roughly 5,000 queries per month.
• The Brainfish team runs internal KB audits, IA cleanup, and coverage testing through MCP. As one knowledge manager put it: "Without MCP I'd have been clicking through the platform manually."

Across the Brainfish customer base, the pattern is the same: knowledge that stays current downstream produces better first answers, fewer escalations, and less content debt.

FAQ

Does Brainfish replace Claude?

No. Brainfish makes Claude more useful for knowledge work by exposing your authored knowledge base via MCP. Claude is still the assistant. Brainfish is the knowledge layer.

Is the Brainfish MCP server official?

Yes. Brainfish builds and maintains the MCP server. It is the first purpose-built MCP server for the authored knowledge layer behind AI customer service.

Does it work with Cursor and Claude Desktop?

Yes. Claude.ai uses OAuth (under 60 seconds). Cursor and Claude Desktop use a JSON config that Brainfish generates for you.

Is my data secure?

Each Brainfish workspace maps to its own MCP tenant. Tokens are scoped, encrypted, and rotatable. Your data stays bound to your tenant and is not used as general training data.

Can Claude actually write back to my knowledge base?

Yes. Drafts created in Claude land in Brainfish as drafts for human review before publish. Audit findings come with reasoning so your team can act with confidence.

How is this different from the Intercom or Zendesk MCP?

Those servers expose ticket and conversation data. Brainfish MCP exposes the authored knowledge layer (search, read, draft, coverage-test). It is the first MCP server built for the knowledge that informs answers, not the conversations after the fact.

What is the Brainfish MCP server URL?

The Brainfish MCP server is at https://mcp.brainfi.sh/mcp. Claude.ai users connect via OAuth at that endpoint in under 60 seconds. Cursor and Claude Desktop users paste the JSON config Brainfish provides into their MCP settings, pointing at the same URL.

What's the fastest way to test it?

Open Claude.ai, run the OAuth flow against https://mcp.brainfi.sh/mcp, and ask: "Find five stale articles in my knowledge base." You'll get a sourced list with reasoning in under a minute.

‍