Quickstart Resources

A repository of servers and clients from the following Model Context Protocol tutorials:

• Quickstart – a simple MCP weather server
• Building MCP clients – an LLM-powered chatbot MCP client

Extended Features

This repository has been enhanced with additional implementations:

Integration Methods

• Claude Integration (index.ts): Original CLI client using Anthropic's Claude
• OpenAI Integration (index-openai.ts): Alternative CLI client using OpenAI GPT models (not for custom GPTs)
• HTTP API Bridge (http-bridge.ts): REST API server for connecting custom GPTs via Actions (recommended for GPT Builder)

Custom GPT Integration

The HTTP API bridge is the only supported method for connecting custom GPTs in OpenAI's GPT Builder:

• REST API server for custom GPT Actions
• File: http-bridge.ts
• Web-based integration via OpenAPI schema
• Endpoints: /weather/forecast, /weather/alerts

How to Connect Your Custom GPT to Your MCP Server

Step 1: Start Your Servers

1. Start the MCP Weather Server

   cd weather-server-typescript
   npm run build
   node build/index.js
   

2. Start the HTTP Bridge

   cd mcp-client-typescript
   npm run build-bridge
   node build/http-bridge.js ../weather-server-typescript/build/index.js
   

Step 2: Expose Your Local Server to the Internet

OpenAI requires a public HTTPS endpoint for Actions. The easiest way is to use ngrok:

ngrok http 3000

• This will give you a public HTTPS URL like https://randomstring.ngrok.io.

Step 3: Prepare Your OpenAPI Schema

• Download or copy your OpenAPI schema from http://localhost:3000/openapi.json.
• Update the "servers" section to use your ngrok URL:

  "servers": [
    { "url": "https://randomstring.ngrok.io" }
  ]
  

• Add a privacy policy and terms of service to the "info" section:

  "info": {
    "title": "Weather API",
    "description": "Weather forecasts and alerts via MCP",
    "version": "1.0.0",
    "termsOfService": "https://www.example.com/terms",
    "x-privacy-policy-url": "https://www.example.com/privacy"
  }
  

• You can use placeholder URLs or your own.

Step 4: Configure Your Custom GPT in GPT Builder

1. Go to ChatGPT GPT Builder
2. Click the Actions tab.
3. Import the schema from your ngrok URL (e.g., https://randomstring.ngrok.io/openapi.json) or paste the edited JSON.
4. Set Authentication to "None".
5. Save and publish your GPT.

Step 5: Test Your Custom GPT

Ask your GPT:

• "What's the weather forecast for San Francisco?"
• "Are there any weather alerts in California?"
• "Get me the forecast for latitude 40.7, longitude -74.0"

Your GPT will call your HTTP bridge, which will call your MCP server and return real weather data!

Using the HTTP Bridge with Remote or Third-Party MCP Servers

By default, the HTTP bridge is designed to launch and connect to a local MCP server (e.g., your weather server). However, you can also use it to connect to any MCP server that is accessible from your machine, including:

• A remote MCP server running on another machine or in the cloud
• A third-party MCP-compatible service
• A Dockerized MCP server with a mapped port

How to Connect the HTTP Bridge to a Remote MCP Server:

1. Ensure the remote MCP server is running and accessible

   • The server must be reachable from the machine running the HTTP bridge (e.g., via public IP, VPN, or SSH tunnel).
   • The server should support the same MCP stdio or TCP transport.

2. Modify the HTTP bridge launch command
   Instead of launching a local script, you can point the bridge to a remote server by:

   • Using an SSH command to start the server remotely and pipe stdio over SSH
   • (Or, for advanced users) Modifying the bridge to use a TCP transport if the MCP server exposes a TCP socket

   Example: Using SSH to connect to a remote MCP server

   node build/http-bridge.js "ssh user@remotehost 'node /path/to/remote/build/index.js'"
   

   • This will use SSH to start the MCP server on the remote host and connect the stdio streams to your bridge.

3. (Optional) Use TCP Transport for Direct Network Connections

   • If your MCP server supports TCP, you can modify the bridge to use TcpClientTransport instead of StdioClientTransport and provide the remote host/port.

   Example (pseudo-code):

   import { TcpClientTransport } from '@modelcontextprotocol/sdk/client/tcp.js';
   // ...
   this.transport = new TcpClientTransport({ host: 'remotehost', port: 12345 });
   

4. Continue with the rest of the setup

   • Expose your HTTP bridge with ngrok or deploy it to a public server as before.
   • Your custom GPT will now be able to access any MCP server you can reach!

Tip:
This flexibility means you can use the HTTP bridge as a universal adapter for any MCP-compatible tool, whether it's running locally, in the cloud, or provided by a third party.

Troubleshooting & Tips

• Public URL Required: OpenAI will not accept localhost or http:// URLs for public GPTs. Always use your ngrok HTTPS URL.
• Privacy Policy: A valid privacy policy URL is required for public GPTs. Use a placeholder or generate one at privacypolicies.com.
• Schema Import Issues: If importing from URL fails, copy the JSON and paste it manually.
• ngrok Free Plan: The URL changes each time you restart ngrok. Update your schema and GPT config if you restart ngrok.
• Production: For a permanent solution, deploy your bridge to a public HTTPS server and use your real domain.

Architecture Overview

Custom GPT ──Actions──> HTTP Bridge ──MCP──> Weather Server ──API──> National Weather Service

Quick Demo

1. Start weather server: cd weather-server-typescript && npm run build && node build/index.js
2. Start HTTP bridge: cd mcp-client-typescript && npm run build-bridge && node build/http-bridge.js ../weather-server-typescript/build/index.js
3. Create custom GPT with Actions pointing to your ngrok URL (see above)
4. Ask: "What's the weather in San Francisco?"

See mcp-client-typescript/README.md for detailed setup instructions.