Skip to main content
Connecting Claude to Cortex lets you ask Claude to remember and recall things across every conversation. This works in Claude on the web (claude.ai) and in the Claude desktop app.

​
What you need

  • A Cortex account (you always have at least your own brain).
  • Claude on a plan that supports custom connectors.
1

Copy your workspace's connection URL

In Cortex, open any brain’s Connect page. Under the connection options, choose the claude.ai tab and copy the URL. It looks like https://api.oncortex.ai/t/your-workspace-id/mcp.The URL identifies your workspace: one connection covers every brain you can access there.
2

Add it as a custom connector in Claude

In Claude, go to Settings, then Connectors, then Add custom connector, and paste the URL. Give it a recognisable name like “Cortex”.
3

Approve the connection

Claude sends you to Cortex to approve. The approval screen names the workspace and shows that Claude will act as you, with your role. Confirm, and you are returned to Claude with the connection live.From here Claude reaches exactly the brains you can reach, at the access you actually hold, checked live on every request.
4

Test it

Ask Claude something like “Using my Cortex connection, what brains can you see?” or “Search my brains for anything about last week’s meeting.” Claude will search and answer with what it finds, saying which brain each thing came from.
Once added, the connector shows the Cortex icon in Claude’s connector list. If yours shows a plain initial instead, it was added before Cortex declared its icon; remove and add it again.

​
For workspace admins: adding Cortex for your whole organisation

A Claude organisation admin can add Cortex once, as a connector for the organisation, so that each member signs in to Cortex from it rather than adding their own. Claude on the web registers itself with Cortex’s sign-in service and keeps a secret Cortex issues it once; each person still approves their own connection in the browser, and Claude reaches exactly the brains that person can. The form asks a handful of things:
  • The address is the workspace connection URL from any brain’s Connect page, the same one shown above.
  • Whether users must sign in: yes, always. Cortex has no anonymous access; every connection acts as a signed-in member.
  • How the connector identifies itself to Cortex: let Claude register itself automatically. There is no client id to paste: Cortex registers the connector on first contact and hands it a secret, which Claude keeps.
  • Authorisation managed by Anthropic: leave it off. Cortex runs its own sign-in and does not read a client description hosted elsewhere.
  • Which tools the connector may use: these can be restricted after the first successful connection, once Claude has seen what Cortex offers.
Members then find Cortex in their connector list, sign in once, and use it exactly as described above.

​
Using it day to day

Once connected, you can ask Claude to both recall and remember:
  • Recall: “What did we decide about pricing?” or “Find my notes on the Acme account.”
  • Remember: “Save this summary to my work brain” or “File these notes in the team handbook.”
  • File a mail thread: if Claude can read your mail through one of its own connectors, “file this thread in Cortex” writes what mattered onto the people and projects it touched and keeps only a link to the message. It goes into the brain marked as yours unless you name another, and only into a brain whose forwarders include the mailbox it came from; before the first thread it files into a shared or client brain in a session, Claude says which brain and asks. See your email.
When Claude files something, it names the destination brain and follows the same skills everything else does, so its filing stays consistent with the rest of your brains.

​
Managing the connection

The connection appears under Connected Agents on the Connect page. To disconnect, revoke it there, or remove the connector in Claude’s settings. Revoking on the Cortex side takes effect immediately.