Brainfish MCP vs Claude Knowledge Base Connectors
May 14, 2026
TL;DR
Guru, Notion, and Confluence are all available as Claude connectors — Claude can read content from them in a conversation. What Claude cannot do, using those connectors, is write back. Creating articles, drafting updates, flagging duplicates, running an audit — none of that is supported. Brainfish MCP is the first bidirectional knowledge layer built for CX teams. Claude reads and writes: search, create, update, draft, audit, coverage-test.
This page uses “read-only vs read-write” in a very specific way:
- Read-only connector: Claude can retrieve and cite content, but cannot create or update the underlying pages/articles from the chat experience.
- Bidirectional (read-write): Claude can retrieve content and create/update knowledge artifacts through tools (for example via MCP), so the same workflow can surface an issue and then draft or apply a fix.</aside>
At a glance
Note on scope: “Claude connector” here refers to the connector experience inside Claude (read/retrieve). Writing changes back into the source system typically requires a separate write-capable integration (for example, an MCP server with create/update tools) and the right permissions/workflows.
Read-only vs read-write is the workflow gap
A read-only KB connector helps Claude answer questions. It does not help your team maintain the KB. If an article is stale, Claude can still cite it. If two articles contradict, Claude may pick one. If there’s a gap, Claude may generate a “best guess” response.
A read-write approach — where Claude can propose and apply changes via tools — lets one workflow both surface the issue and draft the fix. Example: “Is this article still accurate?” → “Draft an update and open it as a PR/review.”
Capabilities: what “write back” actually means
If you’re evaluating “Can Claude do X with our knowledge base?”, these are the practical questions that determine whether it’s just retrieval or a real ops workflow:
- Create: Can Claude draft a new article/page and save it into the system of record?
- Update: Can Claude apply an edit (or create a reviewed draft) on an existing page?
- Audit: Can Claude scan a set of pages for staleness, duplication, contradictions, and coverage gaps?
- Workflow: Can changes flow into approvals, assignments, and revision history instead of silently overwriting?
Choose Brainfish MCP when…
- You want Claude to operate on your knowledge, not just read from it.
- Your KB is a CX asset — deflection rates, agent copilots, help center experience matter.
- You want one knowledge layer Claude operates over content that originates in Confluence, Notion, Google Drive, Guru, or Helpjuice.
- You need reasoning tools (
generate_answer,generate_follow_ups) as first-class MCP endpoints.
Stick with Guru / Notion / Confluence connectors when…
- Your knowledge base is an internal wiki, not a CX asset.
- Read-only access is enough — you don't need Claude to write back.
- You don't need help-desk integrations, deflection widgets, or live agent handoff.
- You're already deeply invested in one of those tools as your source of truth.
Use them together when…
- Your Confluence or Notion is upstream of Brainfish. Brainfish syncs from both. Claude can then read from Brainfish (and write back) while still pulling canonical engineering docs from the source tool when needed.
FAQ
Can Claude edit Notion pages via the Claude Notion connector?
In general, Claude connectors are designed for retrieval (Claude reads/cites content). Editing Notion content usually happens in Notion itself or through a separate write-capable integration/workflow.
Can Claude edit Confluence pages via the Claude Confluence connector?
Typically, connector access is used for reading Confluence content in a conversation. Writing changes back generally requires a dedicated integration/tooling path with write permissions.
Can Claude edit Guru cards via the Guru Claude connector?
Connector-style access is generally read-focused. To create or update Guru content from an AI workflow, you usually need an integration that explicitly supports create/update operations.
Why does bidirectional matter for CX knowledge?
For an internal wiki, “read-only” can be enough. For CX knowledge, the work is continuous: new issues appear, policies change, and articles drift. A bidirectional workflow lets the same conversation move from answering → updating the source → routing for review.
Where can I verify what a connector can do?
Start with the Anthropic Connectors directory listing and the vendor’s documentation for the connector/integration you’re evaluating. (Connector capabilities and scopes can change over time.)
Can I use Brainfish as my knowledge system of record?
Yes. Brainfish can be a stand-alone KB, or it can sync from Confluence, Notion, Google Drive, Guru, Helpjuice, ReadMe, Mintlify, OpenAPI specs, and public websites.
Ready to try Brainfish MCP?