Quickstart Resources

What is Quickstart Resources?

A repository of servers and clients from the Model Context Protocol tutorials, including a simple MCP weather server and an LLM-powered chatbot MCP client. It is intended for developers learning MCP or building custom GPT integrations.

How to use Quickstart Resources?

Build the weather server (npm run build in weather-server-typescript), then start the HTTP bridge (npm run build-bridge in mcp-client-typescript) to connect custom GPTs via OpenAI Actions. Expose the bridge using ngrok and import the OpenAPI schema into GPT Builder.

Key features of Quickstart Resources

• Simple MCP weather server with forecast and alert endpoints
• LLM-powered chatbot MCP client (Claude or OpenAI)
• HTTP API bridge for custom GPT integration via Actions
• Support for remote MCP servers over SSH or TCP
• Step-by-step guide to connect with GPT Builder

Use cases of Quickstart Resources

• Building a custom GPT that queries real-time weather forecasts and alerts
• Demonstrating client-server communication in the Model Context Protocol
• Integrating MCP servers with OpenAI’s GPT Builder for web-based tools
• Testing MCP connectivity across local and remote environments

FAQ from Quickstart Resources

What integration methods are available?

Three methods: Claude CLI (index.ts), OpenAI CLI (index-openai.ts), and HTTP bridge (http-bridge.ts). Only the HTTP bridge works with custom GPTs.

How do I connect a custom GPT to my MCP server?

Start the weather server and HTTP bridge, expose the bridge with ngrok, import the OpenAPI schema into GPT Builder’s Actions tab, and set authentication to None.

Can I use a remote MCP server with the HTTP bridge?

Yes. Use SSH to pipe a remote server’s stdio, or modify the bridge to use TCP transport for direct network connections.

Why does OpenAI require a public HTTPS URL for Actions?

OpenAI’s GPT Builder does not accept localhost or http:// URLs. Use ngrok or a public HTTPS server to expose the bridge.

What dependencies are needed to run the repository?

Node.js and npm are required. For local testing with custom GPTs, ngrok is recommended (free plan works but URL changes on restart).