Skip to navigation
Get started

Get started with Fabric Gateway

This topic describes the product direction and implementation, but much of the platform is still evolving. Some details, capabilities, and decisions may change as development continues.

This topic walks you through connecting your first LLM provider or MCP server and sending a test request through the gateway.

Prerequisites

Before you begin, make sure you have the following:

  • Access to a Fabric AI Gateway organization, with permission to create connections.
  • An API key or token from the provider you want to connect.
  • Postman, if you want to send the test request from a Postman collection.

Connect a provider

You can connect an LLM provider or an MCP server. This topic uses an LLM provider. To connect an MCP server, see Connect an MCP server.

To connect a provider, do the following:

  1. In the sidebar, select LLMs (or MCP servers), and then select the Providers tab.
  2. Find your provider, and then select Connect.
  3. Follow the setup wizard. When it finishes, you have a connection that holds your provider credential and a route that your clients can call. The wizard also creates a gateway key for the route.

Have the following ready before you start:

  • Your provider API key.
  • The models that you want to expose through the route.

For a description of the connection settings, see Connect an LLM provider.

Send a test request from Postman

Before you point an application at the gateway, test the route to confirm that it works. To send a request to an OpenAI route from Postman, do the following:

  1. In the sidebar, select LLMs > Routes, find your OpenAI route, and then select Copy LLM base URL.

  2. In Postman, create a request with the method POST. For the URL, paste the base URL that you copied, followed by /v1/chat/completions.

  3. In the Headers tab, add fg-api-key with your gateway key as the value. For more information, see Authenticate requests with Key Auth.

  4. In the Body tab, select raw and JSON, and then enter the following body:

    {
      "model": "gpt-4o-mini",
      "messages": [
        { "role": "user", "content": "Hello from Fabric Gateway!" }
      ]
    }

  5. Select Send.

The request doesn’t include your provider’s API key, because the gateway adds the key that you saved on the connection.

If you run your own data plane, the copied URL has no host until an admin sets one in Admin > Data plane.

The response is a standard OpenAI chat completion (200 OK) with the model name and the assistant’s reply. It confirms that the gateway proxies the request to the upstream provider while it enforces the gateway key.

If the request returns 401 Unauthorized, see the troubleshooting steps in the Key Auth topic.

Next steps