Setup

Before you install the frontend or backend blocks, read the official ChatGPT Apps SDK documentation. It covers the platform concepts, hosting requirements, and deployment workflow this example builds on.

What You Get

This example ships a single MCP tool named add. The tool:

• Accepts two numbers as input through the MCP request.
• Returns the structured result { a, b, sum }.
• Triggers the accompanying widget to render the inputs and their sum inside ChatGPT.

Once the frontend build script runs and the backend starts, ChatGPT will display a card that shows both operands and their computed sum when the add tool is invoked.

End-to-End Flow

Only personal paid ChatGPT Plans work. Business Accounts don't work yet

Github Issue

1. Install the frontend block. Add the chatgpt-app frontend to your project so you get index.jsx, the widget hooks, and build-chatgpt-widgets.ts.

2. Build the widget assets. Run the build script to emit hashed HTML, CSS, and JS bundles under your configured assets/ directory:

   pnpm tsx build-chatgpt-widgets.ts

   Re-run this command whenever you change the widget UI.

3. Install the backend block. Add the chatgpt-app-backend files so you can host mcp-server.ts, register the add tool, and serve the widget HTML.

4. Expose the MCP server. Start mcp-server.ts, then use ngrok (or a similar tunnel) to forward the port:

   pnpm tsx mcp-server.ts
   ngrok http 8000

   Copy the HTTPS ngrok URL and note the /mcp SSE endpoint.

5. Connect the app inside ChatGPT.

   • In ChatGPT, open Settings → Apps & Connectors.
   • Enable Developer Mode.
   • Create a new app and set its URL to your ngrok endpoint, for example https://your-ngrok-url/mcp.

6. Test the experience. Ask ChatGPT to use your new app. When the add tool runs, it will prompt for two numbers and render the widget that displays both inputs and their sum.