Connect Wendy to ChatGPT

The Wendy plugin brings device inspection, local simulators, and app development into ChatGPT Desktop. It includes the local MCP gateway, the device workspace, and Wendy's bundled skills. Local devices and simulators do not require a Wendy Cloud account.

Add Wendy to your marketplace

Install the Wendy CLI on the computer where you use ChatGPT Desktop, then run:

wendy mcp setup chatgpt

This adds Wendy to your personal marketplace. You can start before connecting hardware or creating a simulator. Setup preserves your existing marketplace entries and previously granted Wendy permissions.

Finish the installation in ChatGPT:

1. Quit and reopen ChatGPT Desktop. Closing a window is not enough.

2. Open Plugins and select Personal, your personal marketplace.

3. Open Wendy and select the plus button to install it.

4. Start a new conversation with Wendy enabled. Ask:

   Help me get started with my first device or a simulator.

ChatGPT launches the local MCP gateway when you use the plugin. The setup command makes the plugin available in your marketplace; the installation step enables it in ChatGPT. This local connection is for Desktop. ChatGPT on the web or mobile needs a separately registered hosted connection.

Connect a device or enable simulators

Default setup enables the device workspace and preferences. Choose the access needed for your task, run setup again, then restart the Wendy connection and start a new conversation.

For a physical device, find its selector with wendy discover --json, then pass the selector to setup. For example:

wendy mcp setup chatgpt --device workshop.local:50052

Repeat --device for multiple devices. Devices added this way allow inspection and installed-app listings. Physical-device camera access and app control require separate permissions in the gateway policy.

For local simulators:

wendy mcp setup chatgpt --simulators

Open Wendy, choose Simulators, then select or create a simulator. Use View simulation to open a supported robot scene. Setup itself does not create or boot a simulator. Simulator access permits lifecycle management, simulator app control, and camera access; it does not extend those permissions to physical devices.

To let ChatGPT read and edit a project and deploy it to permitted simulators:

wendy mcp setup chatgpt --simulators --workspace /path/to/project

Use an existing project directory. For deployment to a physical device, include its --device selector with --workspace. Re-run setup with the same workspace when adding deployment targets.

Optional development and installation tools

Use --host-operations for local installation planning and jobs:

wendy mcp setup chatgpt --host-operations

Disk erases still require authorization for the exact target.

The device workspace already uses the scoped local MCP gateway. Add --developer-tools only when you also want the broader developer MCP server:

wendy mcp setup chatgpt --developer-tools

That additional server has its own target and permissions. It does not inherit the workspace's restrictions. Disable it with --developer-tools=false.

If Wendy does not appear

• Check wendy mcp setup chatgpt --help. If the command is missing, update to a Wendy CLI version that includes ChatGPT setup.
• Run setup on the same computer and user account as ChatGPT Desktop.
• Quit ChatGPT completely and reopen it, then check Plugins → Personal.
• After updating Wendy, restart its MCP connection, refresh its tools, and start a new conversation so ChatGPT loads the current tool catalog and view.

For configuration troubleshooting, setup stores its local policy at ~/.wendy/chatgpt/gateway.json, the plugin at ~/.codex/plugins/sources/wendy-robots, and the personal marketplace entry in ~/.agents/plugins/marketplace.json.

See OpenAI's plugin installation guide for the Plugins Directory and marketplace interface.