By the end of this guide, Claude has access to your notes, documents and scans in Memexa and gives a citation with every answer. You need one URL and about three minutes.
What MCP is
MCP (Model Context Protocol) is the open standard Claude uses to talk to external services. An MCP server provides Claude with tools such as “search my notes” or “read this document”, and Claude calls them on its own when a question fits. In Claude’s interface, such servers are called connectors; you add Memexa as a custom connector. More background is in the MCP fundamentals article; this paragraph is enough for setup.
Prerequisites
- A Memexa account (Basic is enough; 14-day trial, a card is required)
- Claude in the browser (claude.ai), Claude Desktop or Claude Code
A custom connector works in claude.ai and Claude Desktop even on the free Claude plan, though only exactly one; the paid plans allow more. Claude Code needs Pro or above, or a Console account billed by API usage. What each plan costs is in Claude pricing in euros.
You don’t need API keys or config files. Sign-in happens through OAuth in the browser, and the address is the same for every AI app:
https://brain.memexa.ai/mcp
Connect claude.ai and Claude Desktop
The path is the same in the browser and in the desktop app. If you’re signed in to claude.ai, this direct link opens the form straight away; then continue at step 4.
- Open Settings in Claude.
- Choose Connectors.
- Click Add custom connector.
- Enter a name such as “Memexa”, paste
https://brain.memexa.ai/mcpas the URL and confirm. - Sign in with your Memexa account in the browser window that opens. The sign-in is permanent.
Afterwards, the connector appears in your list.
Connect Claude Code
One command in the terminal:
claude mcp add --transport http --scope user memexa https://brain.memexa.ai/mcp
--scope user makes the connection available in all projects, not just the current folder. On first access the browser opens for sign-in. After that the connection runs in the background.
Check that it works
Ask Claude a question only your memory can answer:
“What do you know about me?”
If Claude calls tools like search_brain or list_recent and answers with content from your Memexa, the connection is up. In Claude Code, /mcp shows the status of every connected server.
What Claude can do afterwards
Claude now remembers things across sessions: say “Remember: release moved to the 14th” and the note is still there next week, in ChatGPT or Cursor too if you connect those as well. Uploaded PDFs and scans go through text recognition automatically; ask Claude about one and the answer gives the citation, such as “contract, page 4”. New notes also link themselves to existing ones, so Claude can navigate the graph from one piece of information to the next.
Common problems
The connector shows up, but Claude doesn’t use it
Claude only calls tools when the question sounds like it needs them. Be explicit at first: “Check my Memexa for …”. After the first few hits, Claude starts using the memory unprompted.
I can’t add a second connector
The free Claude plan allows exactly one custom connector. If that slot is taken, remove the old connector, move to Pro or Max, or use Claude Code.
Sign-in fails or hangs
Usually a browser issue: disable the pop-up blocker, then remove and re-add the connector. There is nothing to enter in Claude itself; sign-in happens entirely in the browser.
“Server disconnected” in Claude Code
Run claude mcp list. If memexa shows “failed”, claude mcp remove memexa followed by the claude mcp add from above almost always fixes it. If the error persists, check whether a corporate firewall filters outgoing HTTPS connections.
Claude can’t find a document
Freshly uploaded documents need a moment for text recognition and linking. You can see the status in the Overview. Once the document is marked as processed, Claude finds it.
Other AI apps
To add ChatGPT so both apps read the same memory, see Give ChatGPT and Claude the same memory. For Mistral Vibe and Cursor, the routes are short and complete on How it works: connect AI apps. The address is the same everywhere, and every connected AI app reads the same memory.