🚀 MCP-POC — GPT + Model Context Protocol (MCP) in Action

This is a proof-of-concept implementation demonstrating how to use Model Context Protocol (MCP) with OpenAI function calling to build real-time LLM-native workflows.

---

🔥 What It Does

• Connects GPT-4 to backend tools via the MCP protocol
• Lets GPT choose the right tool dynamically
• Extracts required arguments automatically
• Calls backend logic securely using StdioClientTransport

---

💳 Example Use Case: Block a Credit Card

With just a natural language prompt:

“Please block my card 1234567890123456 because it was stolen.”

✅ GPT:

• Selects the blockCreditCard tool
• Extracts cardNumber and reason
• Calls the backend tool via MCP
• Returns a real-time response

---

🛠 Tech Stack

• 🔗 Model Context Protocol (MCP)
• 🧠 OpenAI function calling (tool_choice: "auto")
• ⚙️ TypeScript + Node.js
• 📦 Zod schema for input validation
• 🔄 Stdio transport for client/server connection
• 🌱 Lightweight and easy to extend

---

📁 Project Structure

src/
├── client/         # Smart OpenAI-powered client
│   └── client.ts
├── server/         # MCP server exposing tools
│   └── index.ts
.env                # Your OpenAI API key
package.json
tsconfig.json

---

🧪 Getting Started

1. Install dependencies

npm install

2. Set your OpenAI key

Create a .env file:

OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

3. Run the project

In one terminal:

npm run start:server

In another terminal:

npm run start:client

4. Try a natural prompt

Please block my credit card 1234567890123456 due to fraud

---

📌 Highlights

• ✅ No hardcoded logic or prompts
• ✅ Tool selection + argument extraction are automatic
• ✅ Works with .ts, .js, or .py server tools
• ✅ Protocol-driven, agent-ready foundation

---

📎 Related

• 🔗 Model Context Protocol Docs
• 📘 OpenAI Tool Calling

---

🙌 Author

Built by Venkata Sairam Gollamudi
📬 vsairamtech@gmail.com