Show HN: OpenAI ChatGPT App starter DevXP feels like 2010, I built a better one (github.com)

A developer published a minimal TypeScript starter that demonstrates how to build an OpenAI Apps SDK–compatible MCP (Model Context Protocol) server with native ChatGPT widget rendering and tool/resource endpoints. The repo combines an Express backend (MCP endpoint at /mcp) with a Vite-powered React frontend (web/src/widgets/) so you can expose callable tools and responses that render inside ChatGPT. It supports Node.js 22+, pnpm, ngrok for local tunneling, and a one-command dev flow (pnpm dev) that runs the server on port 3000 and lets you register the connector in ChatGPT’s Developer mode. Widgets update live via Vite HMR so UI changes appear in an ongoing ChatGPT conversation without restarting the server; server-side changes may require reloading the connector. Why it matters: this template significantly improves developer experience for building ChatGPT Apps by showing a practical MCP implementation, enforcing a clear convention (endpoint name must match widget file name) for automatic mapping, and including both UI and non-UI tools. It also provides a quick path to production via Alpic deployment. For AI/ML engineers and tool builders, this lowers friction for iterating on integrated tools and rich UI components directly within conversations, accelerating prototyping and deployment of model-augmented applications.