> For the complete documentation index, see [llms.txt](https://docs.kula.digital/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.kula.digital/connect-claude-and-access/claude.md).

# Connect Claude

Kula controls access by **rules and roles** — so you never point Claude at Kula directly. Instead there's a **personal connect link**, created in **console.kula.digital**, that already carries exactly what that person is allowed to see. Adding it to Claude is what connects them. There is no open, direct address to connect to; the connect link is the controlled door in.

There are two ways to add it, and **both end up in the same place**:

* **One click** — the invite has an **Open in Claude** button that opens Claude with the connector already filled in. You just review and approve.
* **By hand** — the same screen also shows the connect link, so you can copy it into Claude yourself.

## If someone sent you an invite

The fastest path. You'll have an email (or a link to the connect page) with a big **Open in Claude →** button.

1. Click **Open in Claude →**. Claude opens — the desktop app or claude.ai — with its **Add custom connector** dialog already filled in: the connector name (`Kula — <your studio>`) and the **Remote MCP server URL** (your connect link). Nothing to type, nothing to paste.
2. Review what it says and click **Add**, then **Connect**.
3. **Approve** the connection on the consent screen — it shows the studio's name and the access level being granted.

You're returned to Claude and the connector shows as connected. Skip to [Check it worked](#check-it-worked).

> **Prefer to do it by hand?** You don't lose that option. The email and the connect page both still show your connect link in full, with the manual steps underneath — see [Add it by hand](#add-it-by-hand). Nothing about the one-click button changes what's being granted; it fills in the same two fields you'd type yourself.

The invite also tells you two things worth noting: whether the link is **single-use** (spent the moment it's connected) or **reusable**, and the date it **expires**. If it lapses before you get to it, ask whoever sent it to hit **Extend** — the link you were already sent keeps working, it just gets a new deadline.

## If you're creating the link yourself

1. Sign in to **console.kula.digital**.
2. Open **Connectors** (*"Claude connector invites"*) and click **New invite →**. The **permission level** you choose decides what the connection can access — see [Permission levels](/connect-claude-and-access/scopes.md).
3. You now have a connect link that looks like:

   ```
   https://mcp.kula.digital/connect/XXXXXXXX/mcp
   ```

   Treat it like a password — it's scoped to a role, but it's still a credential.
4. Either **email it to them** from that screen — they get the invite with the **Open in Claude** button — or copy the link and add it to Claude yourself with the steps below.

> Why a link and not an open address? Because access is the whole point. Every connection runs through a role-scoped link, so what each person can see stays under your control. We don't allow direct access to the data.
>
> Connecting Claude on **your own machine**, or using a paste-a-token client like Cursor or Lovable? Mint a **bearer token** under **Tokens** instead — connector invites are for sending access to someone else. See [Connect — API, Code, Cursor, Lovable](/for-developers/connect.md).

## Add it by hand

1. In **Claude Desktop** or **claude.ai**, open **Settings → Connectors → Add custom connector**.
2. Fill in two fields:
   * **Name:** `Kula Intelligence`
   * **Remote MCP server URL:** your **connect link** (`https://mcp.kula.digital/connect/…/mcp`).
3. **Leave the Advanced settings (OAuth Client ID / Secret) blank** — Claude registers itself automatically; filling those in breaks the connection.
4. Click **Add**, then **Connect**. A browser window opens.
5. **Sign in and approve** the connection on the consent screen — it shows the studio's name and the access being granted.

You're returned to Claude and the connector shows as connected.

## Check it worked

Start a new chat and ask:

> *"List the tools you can use from Kula Intelligence."*

Claude should return the full list of tools. If it does, you're connected — and only seeing what your role allows.

For the best answers, pick Claude's most capable model. Then try something real about the studio:

> *"Which members look at risk of churning right now, and what's driving it?"*

## Set up by a consultant?

A consultant helping your studio can create the connect link for you and either hand it over or set it up on your behalf — the access is still yours, scoped to your rules. See [Set up on their behalf](/for-consultants/set-up-on-their-behalf.md).

## Using another assistant?

Claude's custom connector uses the OAuth connect link above. **ChatGPT** can use the same link: open **Plugins** from the sidebar, then the **+** button (or switch the panel to **MCP Servers** → **Add Server**), paste the link as the **Server URL** and set **Authentication** to **OAuth**. The connect page spells out both of ChatGPT's screens. Then **tell ChatGPT to use it** — it won't reach for a new connector on its own. In a new chat, send *"Connect to Kula MCP"*, then *"Run the skill Lets Get Going"*. **Other AI clients** — Cursor, Lovable, the Claude API, custom agents — connect with a **role-scoped access token** instead: you create one in console.kula.digital and it generates the ready-to-paste config for your platform. Same access control, different mechanism. See [Connect — API, Code, Cursor, Lovable](/for-developers/connect.md).

## Trouble connecting?

* **The Open in Claude button didn't fill anything in** → you may have landed in Claude without being signed in, or on a version that doesn't support pre-filled connectors. Sign in to Claude and click the button again — or just [add it by hand](#add-it-by-hand) with the link shown on the same screen. Both do exactly the same thing.
* **"401" / "invalid token"** → your connect link is expired or has already been used. Ask for a fresh one — or, if it's just expired, for an **Extend** on the one you have.
* **"403" / "scope … cannot …"** → the action needs more access than your role allows. See [Permission levels](/connect-claude-and-access/scopes.md).
* More in [Troubleshooting](/connect-claude-and-access/troubleshooting.md).