# Connect Your Assistant

> Connect Claude, Cursor, or any MCP client to your Divinci account — no code, no API key. Sign in with your browser and your assistant can act on your account.

This is the no-code path. Point an MCP-compatible AI assistant — Claude Desktop,
Claude on the web, Cursor, or any other MCP client — at Divinci's hosted server,
sign in once in your browser, and the assistant can search your knowledge base,
read and send chats, and operate on the rest of your account through the
[tool catalog](/mcp/tool-catalog).

You do **not** need the `@divinci-ai/mcp` package, an API key, or any code for
this. (If you're building your own agent in TypeScript, use the
[SDK client](/mcp/overview) instead.)

## Server URL

Divinci runs the MCP server at one endpoint per environment. Use **production**
unless you've been told otherwise.

| Environment | Server URL (Streamable HTTP) |
|-------------|------------------------------|
| Production  | `https://mcp.divinci.app/mcp`       |
| Staging     | `https://mcp.stage.divinci.app/mcp` |
| Dev         | `https://mcp.dev.divinci.app/mcp`   |

<Aside type="caution">
  **Always use the `/mcp` (Streamable HTTP) endpoint.** It's the modern MCP
  transport and the only one that connects reliably to the Divinci server
  (which runs on Cloudflare Workers, where the legacy `/sse` transport's
  in-memory sessions are unreliable). The `/sse` endpoint exists only for
  backward compatibility — if a client defaults to or appends `/sse`, change
  the URL to end in `/mcp`.
</Aside>

## Add the server to your client

<Tabs>
<TabItem label="Claude (web)">

In Claude on the web, open **Settings → Connectors → Add custom connector**,
then paste the server URL:

```
https://mcp.divinci.app/mcp
```

Claude registers itself automatically (Dynamic Client Registration) and opens a
Divinci sign-in window. After you approve, the Divinci tools appear in the
connectors menu.

</TabItem>
<TabItem label="Claude Desktop">

Open **Settings → Developer → Edit Config** (or edit
`claude_desktop_config.json` directly) and add:

```json
{
  "mcpServers": {
    "divinci": {
      "url": "https://mcp.divinci.app/mcp"
    }
  }
}
```

Restart Claude Desktop. On first use it opens a browser window for Divinci
sign-in.

<Aside type="note">
  Older Claude Desktop builds can't reach a remote URL directly. If `"url"`
  isn't recognized, use the `mcp-remote` bridge instead:

  ```json
  {
    "mcpServers": {
      "divinci": {
        "command": "npx",
        "args": ["-y", "mcp-remote", "https://mcp.divinci.app/mcp"]
      }
    }
  }
  ```
</Aside>

</TabItem>
<TabItem label="Cursor">

Add to your Cursor MCP settings (`~/.cursor/mcp.json` or
**Settings → MCP → Add new server**):

```json
{
  "mcpServers": {
    "divinci": {
      "url": "https://mcp.divinci.app/mcp"
    }
  }
}
```

Cursor prompts you to sign in to Divinci the first time it connects.

</TabItem>
<TabItem label="Other clients">

Any MCP client that supports a remote server over Streamable HTTP or SSE works.
Give it the server URL and let it handle OAuth:

```json
{
  "mcpServers": {
    "divinci": {
      "url": "https://mcp.divinci.app/mcp"
    }
  }
}
```

The server advertises OAuth metadata at
`/.well-known/oauth-protected-resource` and
`/.well-known/oauth-authorization-server`, supports Dynamic Client
Registration at `/register`, and exposes an agent card at
`/.well-known/agent-card.json` — so compliant clients discover and
complete the sign-in flow with no manual configuration.

</TabItem>
</Tabs>

## Sign in (the first connection)

<Steps>

1. **Connect.** Your client opens a Divinci sign-in window the first time it
   reaches the server.

2. **Authorize.** Sign in with your Divinci (Auth0) account and approve access.
   The assistant connects on behalf of *your* identity, so it can only see and
   change what your account is permitted to.

3. **Use the tools.** The assistant now lists Divinci tools and can call them.
   Try: *"Search my Divinci knowledge base for refund policy"* or *"List my
   Divinci releases."*

</Steps>

There is **no API key in the config** — authentication is OAuth 2.0 (Auth0,
PKCE). The browser flow runs on first connect and the client refreshes tokens
silently after that.

## Static API key (alternative)

If your client can't do an interactive browser flow — a headless agent, a CI
job — use a Bearer API key from the **API Keys** page of your Divinci workspace
instead of OAuth:

```json
{
  "mcpServers": {
    "divinci": {
      "url": "https://mcp.divinci.app/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_DIVINCI_API_KEY"
      }
    }
  }
}
```

The API key's permissions scope exactly what the assistant can do — the same
permission model as the OAuth identity.

<Aside type="caution">
  Treat the API key like a password. It carries every permission granted to it;
  store it in your client's secret store, never in a shared or committed config
  file.
</Aside>

## What can the assistant do?

Everything your account is permitted to — knowledge-base search, chat,
releases, RAG ingestion, fine-tuning, QA, analytics, notifications, and more.
See the full list:

<CardGrid>
  <LinkCard title="Tool Catalog" href="/mcp/tool-catalog" description="Every tool an assistant can call, grouped by area." />
  <LinkCard title="Authentication" href="/mcp/authentication" description="The Auth0 PKCE flow in detail (for SDK builders)." />
  <LinkCard title="Premium tools & payments" href="/mcp/payments" description="Pay-per-call tools via x402." />
</CardGrid>