brain-sync Sign in

brainsync.fabbrica.co

La Fabbrica’s shared brain

The knowledge your team chose to share, searchable from your own AI assistant. It is split into vaults, and you see only the ones you are granted.

Sign in Connect your assistant

How it works

Curated, never crawled

Nothing is shared behind anyone’s back. Every page takes the same four steps.

  1. 01

    Pick

    A colleague picks notes in their brain-sync. Notes in your Personal folder can never be shared.

  2. 02

    Preview and confirm

    They see the exact text and the target vault, then confirm. Every time.

  3. 03

    Live, with who shared it

    It goes live with who shared it and its source. Sensitive vaults can require a reviewer first; text that reads like instructions to an AI is always held for review.

  4. 04

    Your assistant answers

    Everyone’s assistant answers only from the vaults that person may read, citing the vault, the author and the source.

Vaults and access

Not one big brain: vaults

Each team’s knowledge sits in its own vault. Access is granted per person, per vault.

Approved by an admin

Signing in is not enough: nobody gets access until an admin approves them.

Granted per person

  • read your assistant can search it
  • submit you can share into it
  • review you approve what others share there

New here? Request access

Sign in and send a request. An admin approves it and picks your vaults, or declines it.

Changes apply at once

When an admin removes a grant or a person, or revokes a connected app, it applies on the next request.

Set up

Ready in three steps

  1. Get access

    Sign in with your company Google account or your email and password. If you are new, you will be asked to request access; an admin approves it and picks your vaults.

    Email and password: open the invite link an admin sends you and choose your password.

    Sign in

  2. Connect your assistant

    Add this address as a custom connector or remote MCP server, then sign in in the browser. You never paste a password or token into a chat.

    MCP server address
    https://brainsync.fabbrica.co/mcp

    Claude (claude.ai, desktop and mobile)

    1. Open Settings, then Connectors.
    2. Add a custom connector and paste the address.
    3. Select Connect and sign in.

    On Team and Enterprise plans an owner may have added it for everyone already.

    Claude Code

    1. Run this in a terminal:
    claude mcp add --transport http brain https://brainsync.fabbrica.co/mcp
    1. In Claude Code, run /mcp, pick brain and choose Authenticate.
    2. Sign in in the browser that opens.

    ChatGPT

    1. Open Settings, then Apps and connectors.
    2. Create a connector with the address. Your workspace admin may need to allow custom connectors, or may have added it for everyone.
    3. Sign in when asked.

    Cursor

    1. Open Cursor Settings, then MCP, and add a new server; or put this in ~/.cursor/mcp.json:
    {
      "mcpServers": {
        "brain": {
          "url": "https://brainsync.fabbrica.co/mcp"
        }
      }
    }
    1. Sign in when Cursor opens the browser.

    Other MCP clients

    1. Add a remote MCP server (Streamable HTTP) with the address.
    2. The client opens the sign-in page; sign in.
    3. Check the app name on the approval page before you allow it.

    MCP Inspector: npx @modelcontextprotocol/inspector, transport Streamable HTTP, the address above, then Connect.

  3. Optional: share from your notes

    To share into the brain, install the brain-sync plugin for Claude Code and Codex. You pick the pages, check the exact text and the vault, and confirm.

    Get the brain-sync plugin

Privacy

Said exactly

  • Your company's knowledge base lives on your own server, in your own database.
  • People only get answers from the vaults they have access to; that is checked on your server before anything reaches the assistant.
  • When someone's assistant (Claude, ChatGPT or another) uses the brain, the parts it retrieves are sent to that assistant, as with any AI connector.
  • Approved items and searches are sent to OpenAI to index and search them. Under OpenAI's API terms they are not used to train OpenAI's models.
  • We keep maintenance access so we can update and support it. You can revoke it at any time; we then stop updates and support.