> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agntix.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Connecting

> Server URL, transport, and authentication for the Agntix MCP server.

The Agntix MCP server is a **remote** MCP server. Any client that speaks the Model Context Protocol can connect to it over HTTP and authenticate with an organization API key.

## Endpoint

```text theme={null}
https://mcp.agntix.ai/mcp
```

<Info>
  Confirm the production MCP endpoint for your deployment before publishing. The host follows the same
  convention as the rest of the platform (`api.agntix.ai`, `app.agntix.ai`, `status.agntix.ai`); update
  this value if your environment differs.
</Info>

## Transport

| Transport                         | Use it when                                                                                                                                                                                |
| --------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Streamable HTTP** (recommended) | Your client supports remote MCP servers natively (Cursor, recent Claude clients, custom SDK clients).                                                                                      |
| **stdio bridge**                  | Your client only supports local/`stdio` servers. Run the [`mcp-remote`](https://www.npmjs.com/package/mcp-remote) adapter, which proxies a local stdio server to the remote HTTP endpoint. |

## Authentication

The MCP server uses the **same credentials as the REST API** — see [Authentication](/authentication) for how to create a key.

| Method                    | Header forwarded upstream       | When to use                                     |
| ------------------------- | ------------------------------- | ----------------------------------------------- |
| **API Key** (recommended) | `x-api-key: pk_live_…`          | Server-to-server, local dev, scripts            |
| **Bearer JWT**            | `Authorization: Bearer <token>` | First-party apps with an existing Clerk session |

The scopes on the key apply to every tool call. A `chat:read` key can only invoke read tools; write/delete tools return `403 Forbidden`.

<Warning>
  An API key grants access to your whole organization. Store it in your client's secret/credential
  field — never paste it into a shared config that gets committed to git. Prefer a **scoped, dedicated
  key** for MCP usage.
</Warning>

## Connection config

Most clients accept either a remote URL with headers, or a stdio command. Use whichever your client supports.

<CodeGroup>
  ```json Remote (Streamable HTTP) theme={null}
  {
    "mcpServers": {
      "agntix": {
        "url": "https://mcp.agntix.ai/mcp",
        "headers": {
          "x-api-key": "pk_live_xxxxxxxxxxxxxxxxxxxxxxxx"
        }
      }
    }
  }
  ```

  ```json stdio bridge (mcp-remote) theme={null}
  {
    "mcpServers": {
      "agntix": {
        "command": "npx",
        "args": [
          "-y",
          "mcp-remote",
          "https://mcp.agntix.ai/mcp",
          "--header",
          "x-api-key:pk_live_xxxxxxxxxxxxxxxxxxxxxxxx"
        ]
      }
    }
  }
  ```
</CodeGroup>

## Verify the connection

Once configured, ask your assistant to list its Agntix tools, or have it run a read-only action such as *"list my Agntix agents."* A successful response confirms the server URL and key are correct. A `401` means the key is missing or invalid; a `403` means the key lacks the scope for that tool.

See the [Client Setup](/mcp/clients) page for step-by-step configuration of Claude Desktop, Cursor, and other clients.