Connect Claude to your company documents

Does Claude support MCP servers?

Yes, on every surface. Claude Code, Claude Desktop and claude.ai all take a remote MCP server, and all three reach Tatara at the same address. Claude is the most complete MCP client of the three assistants on this site, so it is the shortest setup.

How do you connect Claude to Tatara?

You give the client Tatara’s address and approve the connection once. Tatara serves MCP over streamable HTTP at a single URL, and takes either a browser sign-in or an access token as the credential, so the steps are the same wherever you start.

• ROUTE 01

  Claude Code and Claude Desktop

  Clients configured by file take the server as one entry.

  {
    "mcpServers": {
      "tatara": {
        "type": "http",
        "url": "https://www.tatara.app/api/mcp"
      }
    }
  }

  1. 1.Add the entry above to your client’s MCP configuration.
  2. 2.The first connection opens your browser. Sign in to Tatara and approve the consent screen, which names the client asking.
  3. 3.The client keeps its own access current from then on, so you do not sign in again every session.
• ROUTE 02

  claude.ai

  Claude on the web takes Tatara as a custom connector.

  1. 1.Open your connector settings and add a custom connector pointing at https://www.tatara.app/api/mcp.
  2. 2.Sign in to Tatara and approve the same consent screen.
  3. 3.On a Team or Enterprise workspace an owner adds the connector once, and each person then connects to it individually.

  Claude on the web connects from Anthropic’s cloud rather than from your machine, so the server has to be reachable on the public internet. Tatara is.

The full procedure, including issuing and revoking access tokens, lives in the Tatara docs. For Claude’s own screens, see Anthropic’s guide to custom connectors.

What can Claude do once it is connected?

It gets twenty-three tools against your brain, in five groups: finding and reading documents, writing them, organising them into folders, curating the tag and property vocabulary, and deleting or restoring things. The tools describe themselves, so a connected agent discovers what it needs when it needs it.

Every agent write is governed. A document needs a type, a title, a description and tags steered to your taxonomy, and a write that does not qualify comes back with what is wrong and how to fix it. The agent corrects itself inside its own loop instead of leaving a half-written document for someone to find later. Every save is stamped with the credential behind it, which is why an agent cannot pass itself off as a person.

Underneath, a brain is plain Markdown with YAML frontmatter: one file per concept, with links forming a graph, and nothing you could not open in an editor. You can also install the open source agent skills with npx skills add tatara-app/tatara-skills, which teach Claude the workflow up front so its first write usually passes.

What are the limits with Claude?

MCP support is not the same across assistants, so here is the honest version for this one.

• Every Tatara connection carries read and write access. There is no read-only credential, so connect Claude only if you are content for it to create and edit documents on your behalf.
• On a Team or Enterprise workspace, adding the connector is an owner action. Members connect to it themselves afterwards, so Claude only ever reaches what that person can reach.
• Disconnecting a browser-signed-in client stops it renewing its access, but the credential it already holds keeps working for up to an hour. An access token is cut off on its very next request.

Claude and your company brain

What does it cost to connect Claude?

Nothing to start. The free plan holds 500 documents and does not ask for a card. Agents connect to it over MCP and can read your brain and make governed writes to it. Paid plans add headroom, and the pricing page has the current limits.