> 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.

# Manage gateway connections and routes

Fabric AI Gateway can manage your **LLM providers**, **MCP servers**, and plain **REST APIs**. They share the same underlying connection model — a connection, one or more routes, and an upstream target — but each has settings specific to its protocol.

Learn how to manage LLM connections and routes, configure an MCP server connection, and connect a REST API as a service. If you haven't connected anything yet, start with the [Quickstart](/get-started/quickstart/).

## Connect and manage LLM providers

**LLMs** has the following components:

| Component       | What it's for                                                                                                    |
| --------------- | ---------------------------------------------------------------------------------------------------------------- |
| **Connections** | Every LLM provider you've connected, such as `anthropic`, `bedrock`, or `openai`.                                |
| **Routes**      | The route generated for each connection, including every path it exposes and the plugin instance attached to it. |
| **Providers**   | The full catalog of LLM providers available to connect in your tenant.                                           |

### View your connections

To view your LLM connections, do the following:

1. In the sidebar, select **LLMs** under **Gateway**.
2. Select the **Connections** tab.

Each connection shows its **target URL**, a short description, tags, and its upstream count. To add another provider, select **Create connection**.

### View your routes

To view the routes generated for your connections, do the following:

1. In the sidebar, select **LLMs** under **Gateway**.
2. Select the **Routes** tab.

Each row shows a route's paths and methods, along with the plugin instance attached to it. For example:

| Route       | Paths                                                                                                      | Plugin instance         |
| ----------- | ---------------------------------------------------------------------------------------------------------- | ----------------------- |
| `anthropic` | `/anthropic/v1/messages` (POST), `/anthropic/v1/messages/count_tokens` (ANY), `/anthropic/v1/models` (GET) | `route-auth--anthropic` |
| `openai`    | Chat completions, responses, embeddings, models, audio transcription/speech, image generation, moderations | `route-auth--openai`    |

### Browse available providers

To browse the LLM provider catalog, do the following:

1. In the sidebar, select **LLMs** under **Gateway**.
2. Select the **Providers** tab.

This tab lists every LLM provider available to connect in your tenant — 22 providers, in a typical tenant.

## Configure an MCP server connection

When **Service type** is set to **MCP**, the connection editor exposes tools and resources to agents over MCP (JSON-RPC/SSE), rather than proxying plain REST or LLM traffic.

### Configure an MCP connection

To configure an MCP connection, do the following:

1. In the sidebar, select **MCP servers** under **Gateway**, then open the connection you want to configure (for example, `linear-mcp`).
2. Under **Service type**, select **MCP**.
3. Enter the **Target URL** for the MCP server, for example `https://mcp.linear.app/mcp`.
4. Under **Upstream auth requirement**, select an **Auth type** (for example, **Bearer**) and set **Applies to** (for example, **Org/tenant**).
5. Optionally, expand **Identity binding** or **Advanced settings** to configure identity mapping.
6. Enter a **Description** and, optionally, select a **Category**.
7. Click **Save changes**.

### Add upstreams

An MCP connection can point to one or more upstreams. Each upstream has its own **Target URL**, **Weight**, and **Name**, along with a **Per-upstream OAuth / MCP discovery** checkbox.

To add an upstream, do the following:

1. In the connection editor, click **Add upstream**.
2. Enter a **Target URL**, **Weight**, and **Name** for the upstream.
3. Select a **Load balancer** strategy.

Weight controls how traffic is split across upstreams, but it's only used by load balancer strategies that support weighted splitting. Under **Round robin**, traffic rotates evenly across upstreams and weight is ignored.

### Set the access mode

Under **Access mode**, select one of the following:

| Access mode | Behavior                                               |
| ----------- | ------------------------------------------------------ |
| **Open**    | The connection is reachable without additional gating. |
| **Gated**   | The connection requires gating before it's reachable.  |

## Connect a REST API as a service

**Services** connects plain HTTP APIs the same way you connect LLM providers and MCP servers. Add a service, then import an OpenAPI spec to generate its routes automatically.

### Import an OpenAPI spec

**Import an OpenAPI spec** is a 3-step wizard.

#### Step 1: Provide spec

Paste or upload a JSON/YAML OpenAPI document. Under **Import mode**, select one of the following:

| Import mode   | Behavior                                                                     |
| ------------- | ---------------------------------------------------------------------------- |
| **Merge**     | Add and update routes from the spec; leave routes not in the spec untouched. |
| **Overwrite** | Match the spec exactly, removing any routes that aren't in it.               |

Optionally, expand **Advanced settings** to set:

* **Import ID** — a stable identifier so re-imports keep updating the same API instead of creating a new one.
* **Route prefix** — mounts every route from the spec under a path prefix, which is stripped before the request is forwarded upstream, so multiple imported APIs don't collide.
* **Upstream URL override**.

#### Step 2: Review plan

Review the plan before you publish anything. The plan shows which service will be created or updated, how many routes are new, changed, or unchanged, and the exact route paths and methods.

Imported services don't get gateway-managed auth by default — callers pass their own credentials through to the upstream unless you add an upstream credential on the **Credentials** page after publishing. For how to add one, see [Manage users, roles, and credentials](/fabric/administration/).

#### Step 3: Publish

Click **Publish** to apply the plan.

### Re-import to update routes

Re-importing an updated spec in **Merge** mode applies incremental updates. For example, adding a `PUT` operation to an existing path and a new path produces a plan showing 1 new, 1 changed, and 3 unchanged routes, with an option to expand and view the unchanged ones.

Routes from multiple imports appear together under **Services → Routes** — for example, two services imported from the same spec at different times can end up contributing a combined 10 routes across `albums`, `comments`, `posts` (with both GET and PUT), `todos`, and `users`.

## Next steps

* To add authentication, rate limiting, or transforms to any of these routes, see [Control and transform gateway traffic with plugins](/fabric/plugins/).
* To monitor traffic on these routes, see [Monitor gateway traffic with analytics and traces](/fabric/observability/).
* To add an upstream credential for a connected service, see [Administration](/fabric/administration/).