ChatGPT Apps SDK EdgeOne Pages Starter

This repository is a minimal EdgeOne Pages project that shows how to build an MCP server compatible with the OpenAI Apps SDK using Next.js and edge functions.

Deploy

Live Demo: https://chatgpt-apps-edgeone-pages.edgeone.run

Overview

The project demonstrates how to host an MCP server with Tencent Cloud EdgeOne Pages + Functions using Next.js. The MCP server exposes tools to ChatGPT and renders widgets with structured content.

Tech Stack

• Next.js 15: React framework with App Router

• Hono: Fast, lightweight web framework for EdgeOne Functions

• MCP SDK: Model Context Protocol implementation

• Tailwind CSS: Utility-first CSS framework

• TypeScript: Type-safe development

Capabilities

• MCP server: A serverless onRequest handler built on EdgeOne Functions using Hono

• Widget support: The hello_stat tool renders dynamic widgets with structured content

• CORS support: Built-in logic keeps the server compatible with ChatGPT iframes and browser debugging

• Next.js frontend: Modern React-based UI with Tailwind CSS styling

Key files

• functions/mcp/index.ts: Implements MCP JSON-RPC, registers the hello_stat tool, and handles widget rendering

• functions/httpTransport.ts: Custom HTTP transport for MCP server

• app/page.tsx: Next.js landing page that explains the project

• app/layout.tsx: Root layout with global styles

• app/globals.css: Tailwind CSS configuration and global styles

• edgeone.json: EdgeOne Pages configuration with CORS headers

Quick start

1. Deploy to EdgeOne Pages

1. Click the button above for one-click deployment.

2. After provisioning, the EdgeOne console assigns a domain that hosts both the Next.js app and the MCP endpoint.

Once deployment finishes, the Next.js app is served from the root path and functions/mcp/index.ts is automatically mapped to /mcp.

Connect from ChatGPT

1. Ensure your account has ChatGPT Apps developer access.

2. In ChatGPT, open Settings → .

3. Add the EdgeOne Pages deployment URL as the MCP server, for example:

   https://<your-project-url>/mcp

4. Save the configuration, then call the hello_stat tool in a conversation to render a stat widget with your name.

MCP request flow

1. ChatGPT calls /mcp via the MCP protocol, triggering the initialize and tools/list handshake.

2. When a user invokes hello_stat, the server returns structured content with title/value/description.

3. The widget template is fetched from the Next.js app root and rendered in ChatGPT with the structured data.

4. Errors are returned in a structured format so ChatGPT and developers can debug quickly.

Project structure

Further reading

• OpenAI Apps SDK docs

• Model Context Protocol spec

• EdgeOne Pages & Functions guide

• ChatGPT connector creation guide

To extend the setup, continue evolving the MCP logic in functions/mcp/index.ts. EdgeOne Pages scales compute as needed.