> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.fabricgateway.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.fabricgateway.ai/_mcp/server.

# Authenticate and send a request

> Find the gateway's base URL, send your gateway key in the fg-api-key header, and make your first request through Fabric AI Gateway.

To send traffic through Fabric AI Gateway, you need two things: the base URL of a route and a gateway key from the route's namespace. You send the gateway key in the `fg-api-key` request header. The gateway key authenticates you to the gateway. If your route uses JWT Auth instead, send the token that your identity provider issued in the `Authorization` header, as `Bearer TOKEN`. For details, see [Authenticate requests with JWT Auth](/configure/authentication/jwt-auth). You don't send the provider's API key, because the gateway adds it when it proxies the request. This topic shows you how to find the base URL and send a request to an OpenAI-compatible route and to an Anthropic route.

## Prerequisites

Before you begin, make sure you have the following:

* A gateway key from the namespace that holds the route. Ask your admin for one, or create one yourself in **Credentials** if your account allows it. For details, see [Authenticate requests with Key Auth](/configure/authentication/key-auth).
* Postman or cURL, to send the request.

> **Warning**
>
> Treat gateway keys as secrets. Don't share them in screenshots, recordings, or chat messages. If a key leaks, rotate it in **Credentials**, or ask your admin to rotate it.

## Get the base URL

Each route has its own base URL. To copy it, in the console sidebar, select **LLMs**, select the **Routes** tab, find the route, and then select the **Copy LLM base URL** icon in its row. The copied URL doesn't end in `/v1`. If the route's only path is a wildcard, such as `/openai/*`, the copied URL keeps the trailing `/*`. Remove it before you append an endpoint. If you don't have access to the console, ask your admin for the URL.

The route's ⋯ menu also offers full URLs and ready-made cURL commands for its endpoints, such as **Copy chat URL (POST)** and **Copy models cURL (GET)**.

The URL's host is the data plane host for your organization. On a Postman-hosted data plane, the host is a subdomain of `dp.fabricgateway.ai`. Copy it from the console rather than typing it. If your organization runs its own data plane, the host is the one your admin set for it.

> **Note**
>
> If your organization runs its own data plane, check that an admin has set its URL in **Admin** > **Data plane**. Until then, the copied URL starts with Postman's hosted gateway address, `https://dp.fabricgateway.ai`, not your data plane's host.

## Send a request to an OpenAI-compatible route

To send a chat completion request, append `/v1/chat/completions` to the base URL. To send a request, choose Postman or cURL.

#### Postman

1. Create a request with the method `POST`. For the URL, paste the base URL that you copied, followed by `/v1/chat/completions`.
2. In the **Headers** tab, add `fg-api-key` with your gateway key as the value.
3. In the **Body** tab, select **raw** and **JSON**, and then enter the following body:

   ```json
   {
     "model": "gpt-4o-mini",
     "messages": [
       { "role": "user", "content": "Hello from Fabric Gateway!" }
     ]
   }
   ```
4. Select **Send**.

#### cURL

Replace `BASE_URL` and `GATEWAY_KEY`:

```bash
curl "BASE_URL/v1/chat/completions" \
  -H "fg-api-key: GATEWAY_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [
      { "role": "user", "content": "Hello from Fabric Gateway!" }
    ]
  }'
```

A successful request returns `200 OK` with a standard OpenAI chat completion: the name of the model that answered and the assistant's reply.

## Send a request to an Anthropic route

An Anthropic route accepts the Anthropic Messages API. To send a request, choose Postman or cURL.

#### Postman

1. Create a request with the method `POST`. For the URL, paste the base URL of your Anthropic route, followed by `/v1/messages`.
2. In the **Headers** tab, add the following request headers:

   | Key                 | Value             |
   | ------------------- | ----------------- |
   | `fg-api-key`        | Your gateway key. |
   | `anthropic-version` | `2023-06-01`      |
3. In the **Body** tab, select **raw** and **JSON**, and then enter the following body:

   ```json
   {
     "model": "claude-sonnet-4-5",
     "max_tokens": 256,
     "messages": [
       { "role": "user", "content": "Hello world!" }
     ]
   }
   ```
4. Select **Send**.

#### cURL

Replace `BASE_URL` and `GATEWAY_KEY`:

```bash
curl "BASE_URL/v1/messages" \
  -H "fg-api-key: GATEWAY_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-5",
    "max_tokens": 256,
    "messages": [
      { "role": "user", "content": "Hello world!" }
    ]
  }'
```

A successful request returns `200 OK`. The response body contains an assistant message in the `content` array, with the name of the model that answered and token counts in the `usage` object. Your request doesn't include the provider's API key, because the gateway adds it.

## Understand how your key picks a namespace

Your gateway key, not the URL, decides which namespace serves your request. A key works only in the namespace that issued it. The gateway handles each request in this order:

1. It reads the `fg-api-key` header. With no key, the request goes to the Default namespace. With a known key, it goes to that key's namespace. With a key that the gateway doesn't know, the request stops with `401`.
2. It looks for a matching route among that namespace's routes only. If nothing matches, you get `404`.

A namespace has no URL form on the data plane. A path such as `/finance-agent-demo/...` is just the path that an admin gave a route. Only the `fg-api-key` header picks the namespace, so a key sent in `Authorization` or `x-api-key` doesn't, and the request goes to the Default namespace.

The following table shows what you get:

| What you send                                                                  | What happens                                                                                                                           |
| ------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------- |
| No key, to a route in the Default namespace that uses Key Auth                 | The route matches, and Key Auth refuses the missing key with `401`. A Default route with no authentication plugin accepts the request. |
| No key, to a route in another namespace                                        | The request goes to Default, which has no route at that path, so you get `404`.                                                        |
| A key from namespace A, to a route in namespace A                              | The route matches, and the request continues.                                                                                          |
| A key from namespace A, to a route in the Default namespace                    | The request goes to namespace A, which has no route at that path, so you get `404`.                                                    |
| A key from namespace A, to a route in namespace B                              | The same: `404`.                                                                                                                       |
| A key the gateway doesn't know                                                 | `401`, before any route is matched.                                                                                                    |
| Two different key values in the headers on the route's Key Auth `headers` list | `401` with `conflicting credentials`.                                                                                                  |
| A key, before the data plane has received its first configuration              | `503` with `service unavailable`. Retry the request.                                                                                   |
| No key, before the data plane has received its first configuration             | `404`, because the data plane has no routes yet. Retry the request later.                                                              |

One exception: an LLM route in the Default namespace that has a model router can also serve other namespaces. For details, see [Serve every namespace from a Default route](/configure/routing/model-router#serve-every-namespace-from-a-default-route).

A URL that you open in a browser can't send the `fg-api-key` header, so it always lands in the Default namespace. To test a route outside Default, use a copied cURL command or an HTTP client that sends the header.

## If the request fails

If the request returns `401 Unauthorized`, see [Troubleshoot authentication errors](/configure/authentication/key-auth#troubleshoot-authentication-errors).

## Next steps

* To use the gateway from a coding agent, see [Use Claude Code with Fabric AI Gateway](/guides/coding-agents/claude-code), [Codex](/guides/coding-agents/codex), or [OpenCode](/guides/coding-agents/opencode).
* To review the request you sent, see [Inspect traces](/use/monitor/traces).