# Connect your assistant to Lorito

Lorito learns a website and turns it into an AI agent that answers questions on a website widget, an embed, a hosted chat page or a custom domain. These instructions are for the AI assistant reading them.

Connect the user to Lorito with the tools you actually have in this conversation. Do the mechanical steps yourself where you can. Ask the user only for a sign-in, a setting only they can change, or a missing detail. Keep progress updates short.

## Connection details

| Setting | Value |
| --- | --- |
| Name | Lorito |
| Server URL | `https://www.lorito.ai/api/mcp` |
| Transport | Streamable HTTP |
| Authentication | OAuth 2.1 in the browser, with dynamic client registration |

Lorito has no CLI and no API keys. The remote MCP server is the only way to connect. Never ask the user for a token or password in chat, and never build an OAuth URL yourself: show only a sign-in link your host returns.

## 1. Reuse an existing connection

Inspect your available tools and connectors, including deferred tools you can load. Lorito tools are named `org_list`, `project_list`, `knowledge_search` and so on, sometimes with a host prefix such as `mcp__lorito__`.

If they exist, verify them as in step 4. If `org_list` fails with 401 or an auth error, use your host's re-authentication flow once, then call it again.

## 2. Connect

Follow the section for your host. Do not ask the user to choose a method. If you cannot tell which host you are running in, ask its name once.

### Claude Code

Run these yourself. First check for an existing entry:

```bash
claude mcp get lorito
```

If it exists and points at `https://www.lorito.ai/api/mcp`, skip the add. If it points anywhere else, run `claude mcp remove lorito` first. Then add Lorito for every project and sign in:

```bash
claude mcp add --transport http --scope user lorito https://www.lorito.ai/api/mcp
claude mcp login lorito
```

`claude mcp login` opens the browser and waits while the user signs in. Without a local browser, add `--no-browser`: it prints the sign-in link for the user. If the Lorito tools do not appear in this session afterwards, ask the user to restart Claude Code and paste the resume prompt from step 4.

### Cursor

Add this server to `~/.cursor/mcp.json` so it works in every project. Merge it into any existing `mcpServers` entry; do not overwrite other servers.

```json
{
  "mcpServers": {
    "lorito": {
      "url": "https://www.lorito.ai/api/mcp"
    }
  }
}
```

If you cannot edit files, give the user this install link instead: `cursor://anysphere.cursor-deeplink/mcp/install?name=lorito&config=eyJ1cmwiOiJodHRwczovL3d3dy5sb3JpdG8uYWkvYXBpL21jcCJ9`. Then ask the user to open **Customize** in the sidebar, check Lorito is toggled on, and sign in when Cursor asks. If the tools still do not appear, ask the user to restart Cursor and paste the resume prompt from step 4.

### Claude.ai and the Claude apps

You cannot add a connector from inside a chat. Give the user these steps:

1. Open **Customize > Connectors**, click **+ Add**, then **Add custom connector**. On a Team or Enterprise plan an Owner adds it first in **Organization settings > Connectors** (**Add**, then **Custom > Web**); members then click **Connect** on it.
2. Name it `Lorito`, paste `https://www.lorito.ai/api/mcp` as the remote MCP server URL and click **Continue**.
3. Review the detected authentication settings and click **Continue**.
4. Under **Authentication** choose **Sign in now**. Under **OAuth client** choose **Register automatically**. Leave **Request headers** empty and click **Add**, then sign in to Lorito.
5. In a new chat, click **+**, open **Connectors** and turn Lorito on.

### ChatGPT

You cannot add a connector from inside a chat. Custom MCP servers work in ChatGPT on the web, and workspace permissions apply. Give the user these steps:

1. Open [Plugins](https://chatgpt.com/plugins), select **+**, then **Add custom MCP server**.
2. Name it `Lorito`. Under **Connection**, paste `https://www.lorito.ai/api/mcp` as the server URL, and choose **OAuth** for authentication.
3. Review the risk warning, select **I understand and want to continue**, then **Create as a plugin**, and install it.
4. Start a new chat, type **@** and pick Lorito. Sign in when asked.

If the user cannot see **Add custom MCP server**, their account or workspace does not allow it. Say so; do not guess at other settings.

### v0

You cannot add a connector from inside a chat. Give the user these steps:

1. In the prompt form, open **+** and select **MCPs**, then add a custom MCP server.
2. Paste `https://www.lorito.ai/api/mcp` as the URL, choose **OAuth** for authentication and sign in to Lorito.
3. Continue this chat or start a new one. v0 uses connected MCP tools automatically.

### Any other host

If your host can add a remote MCP server with a tool, use it with the details above. If the user has to do it in settings, read your host's official instructions and give the smallest exact step; do not invent menu paths. If your host cannot connect remote MCP servers, say so plainly. You can still explain Lorito, and the user can work in the dashboard at https://www.lorito.ai.

## 3. Sign in

Sign-in opens Lorito in the browser. The user signs in to their Lorito account, or creates one, and approves access. Tell the user honestly what approving grants. The consent screen lists it:

- Create and edit projects, agents, and channels
- Add, remove, and search knowledge
- Add and remove skills
- Read your visitor conversations

Across every organisation on your account. It cannot delete projects, manage billing, or change access.

Setup itself only reads. Wait for the user to finish, then check your tools again. Do not restart setup from scratch. If sign-in was cancelled, offer one new attempt.

## 4. Verify and hand over

Call `org_list`. It is read-only and takes no arguments. Every Lorito account has at least a personal organisation, so a valid response lists one or more. An empty list, an error inside the result, a saved config, a connected badge or the user saying "done" is not verification.

Tell the user in one line which organisation names you can see, so they can spot the wrong account. Do not recite other account details.

- **Empty list or wrong account:** the token belongs to another account or is broken. Disconnect Lorito in your host (Claude Code: `claude mcp logout lorito`; elsewhere, disconnect it in the connector settings), sign in again with the right account, then call it again.
- **New chat needed:** if your host only loads new tools in a new chat or after a restart, say so and give the user this resume prompt to paste there: `Read https://www.lorito.ai/agents/setup.md and verify my Lorito connection.`
- **Network or rate-limit error:** retry once after a short wait, then report it.
- Do not create a project or change anything during setup.
- Say whether the connection persists: a saved config or account connector does; a one-off chat session may not.

Once verified, end with exactly:

> You're connected to Lorito. What would you like to build or improve?

Then wait for the answer. Each tool's description explains what it does, so read it before calling. Read the current state before changing anything, and confirm changes to a live agent with the user first. Product docs live at https://www.lorito.ai/docs.
