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

> ## Agent Instructions
> For agents: install the OpenWork desktop app with `brew install --cask openwork` (macOS) or from https://openworklabs.com/download. Do not run `npx openwork`; the npm package named `openwork` is a different project.
> Connect an agent to OpenWork with the MCP Gateway: `claude mcp add --transport http openwork https://api.openworklabs.com/mcp/agent`. Teams sign up at https://app.openworklabs.com?mode=sign-up.
> Step-by-step agent guide: https://openworklabs.com/docs/start-here/use-openwork-from-an-ai-agent.md
> Connect Slack (any organization admin): https://openworklabs.com/docs/slack/connect-slack.md. OpenWork in Slack (@openwork bot) is private alpha for approved organizations. Setup guide: https://openworklabs.com/docs/slack/overview.md. Agent skill: https://openworklabs.com/.well-known/agent-skills/set-up-openwork-slack/SKILL.md.

# Use Claude Desktop with AI Gateway

> Run the Claude app in Gateway mode on your organization's AI Gateway, with an AI Gateway API key and no Anthropic account.

With an [AI Gateway API key](https://openworklabs.com/docs/ai-gateway/third-party-tools.md) you can run Claude
Desktop on the Claude models your organization gives you. Claude Desktop's
**Gateway** mode sends every chat, Cowork task and Claude Code session through
AI Gateway: the key only reaches models you're allowed, and usage counts toward
your limit. You don't need an Anthropic account.

> [!NOTE]
>
> Claude Desktop lists only Claude models in Gateway mode. Your organization's
> other models (OpenAI, Gemini, DeepSeek and the rest) don't appear in its model
> picker. Use them from [Claude Code](https://openworklabs.com/docs/ai-gateway/claude-code.md),
> [Cursor](https://openworklabs.com/docs/ai-gateway/cursor.md) or [Codex](https://openworklabs.com/docs/ai-gateway/codex.md).

## Set it up

Create a key in **My Library › Models** and pick **Claude Desktop** under
**Start using this key**. Already have a key? Open its menu and select **Set
up**. The page shows the two values to paste.

1. **Turn on Developer Mode.** Open Claude Desktop (you don't need to sign in)
   and select **Help › Troubleshooting › Enable Developer Mode**. A
   **Developer** menu appears.
2. **Open the inference settings.** Select **Developer › Configure Third-Party
   Inference…** and set **Connection** to **Gateway**.
3. **Enter the gateway settings:**

   | Field | Value |
   | - | - |
   | **Credential kind** | `Static API key` |
   | **Gateway base URL** | `https://<your AI Gateway>/api/v1/claude-code` |
   | **Gateway API key** | Your AI Gateway API key |
   | **Gateway auth scheme** | `bearer` |

   Leave the sign-in fields (**Client ID**, **Issuer URL**, **Authorization
   URL**) and **Sign-in session lifetime** empty. **Test connection** checks the
   key and lists your models.
4. **Apply.** Select **Apply Changes** (**Apply locally** in older versions),
   then **Relaunch now**, or quit Claude Desktop completely and open it again. If
   the start screen asks, choose **Continue with Gateway** (shown as **Local
   configuration**).
5. **Pick a model.** The model picker under the message box lists your Claude
   models. The account menu at the bottom left shows **Gateway**.

To change the settings later, open the account menu and select **Inference
configuration**.

## Models

Claude Desktop asks AI Gateway for your models when it starts and keeps those
whose name is a Claude model. They have the same names as in
[Claude Code](https://openworklabs.com/docs/ai-gateway/claude-code.md#model-names), for example
`openwork/claude-sonnet-4.5-k3f9`, and show in the picker by their display name.
Models your organization adds later appear after you relaunch Claude Desktop.

## When it doesn't work

| Claude Desktop shows | What to do |
| - | - |
| A connection error, or **Test connection** fails | Check the base URL ends in `/api/v1/claude-code`, the key is pasted in full, and the auth scheme is `bearer`. A revoked or expired key, or 3rd party tools turned off for your organization, gives `authentication_error`. Check the key in **My Library › Models**. |
| An empty model picker | Your organization hasn't given you a Claude model yet. The setup page says which models Claude Desktop will list. |
| **Continue with Gateway** doesn't show | Select **Apply Changes**, then quit Claude Desktop completely (not just its window) and open it again. |
| No **Developer** menu after switching to Gateway | Developer Mode is set separately for Gateway mode. Use the account menu's **Inference configuration**, or turn on Developer Mode again. |
| "Access to this website is blocked by your network egress settings" when Claude fetches a web page | Claude Desktop limits where its tools can connect. Open the inference settings, select **Egress** (**Workspace restrictions** in some versions) and add the hosts to **Allowed egress hosts**. |
| TLS errors from Claude Code inside Claude Desktop | Claude Code's sandbox can break TLS connections to the gateway. Add `{"sandbox": {"enableWeakerNetworkIsolation": true}}` to `~/.claude/settings.json` (every project) or `.claude/settings.local.json` (one project), then relaunch. This relaxes network isolation for all of Claude Code's traffic, so use the narrowest file and remove it when you no longer need it. |

## Go back to your Claude account

Remove the Gateway connection in the inference settings and relaunch. Gateway
mode doesn't touch your Claude account's chats: Claude Desktop keeps Gateway
mode's data separately.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.