# Ecommerce MCP Chat Server Demo with Local LLM

A complete demonstration of connecting a local Large Language Model (LLM) to a NestJS database backend using the **Model Context Protocol (MCP)**. This project allows an AI agent to autonomously execute tools to query a MySQL database in real-time during a chat session.

## 🚀 Features

* **NestJS Backend:** A modular architecture managing Users, Products, and Orders.
* **Database Integration:** Sequelize ORM connected to a local MySQL database.
* **MCP Server:** Exposes database queries natively to AI agents using `@nestjs-mcp/server`.
* **Local AI Agent:** A Node.js CLI chat interface powered by the **Vercel AI SDK** and **Ollama** (`llama3.1:8b`).
* **Agentic Tool Calling:** The LLM autonomously decides when to query the database to answer user questions accurately.

## 📋 Prerequisites

Before you begin, ensure you have the following installed:
* **Node.js** (v24+)
* **MySQL** (Running locally on port 3306)
* **Ollama** (Running locally with the `llama3.1:8b` model pulled)

> **Note:** To pull the required model, run `ollama pull llama3.1:8b` in your terminal.

## 🛠️ Installation & Setup

1. **Clone the repository and install dependencies:**
   ```bash
   npm install
   ```

2. **Database Setup:** Ensure your local MySQL server is running. Create an empty database named **mcp_demo**.

   ```bash
   CREATE DATABASE my_db;
   ```
  (Update the database credentials in src/app.module.ts and src/seed.ts if your MySQL username is not root or if you have a password).

3. **Seed the Database:** Generate dummy data (50 users, 50 products, and 50 orders) to test the AI's querying capabilities.

   ```bash
   npx ts-node src/seed.ts
   ```

## 💻 Running the Application
This project requires two terminal windows to run simultaneously—one for the NestJS MCP Server, and one for the AI Chat Client.

**Terminal 1:** Start the NestJS MCP Server
Start the backend server so it can expose the database tools via the MCP Streamable HTTP transport.

   ```bash
   npm run start
   ```
(The server runs on http://localhost:3000 with MCP available at /mcp)

**Terminal 2:** Start the AI Chat Client
Start the interactive command-line interface. The client will connect to the NestJS server, discover the tools, and allow you to chat with the local LLM.

   ```bash
   npx ts-node src/chat.ts
   ```

## 💬 Usage Examples
Once the chat client is running, try asking the agent questions that require database knowledge:

*"What products do we have available?"*

*"Can you give me the details for user ID 5?"*

*"How many orders are in the system?"*

*"Check the price of product ID 12 and tell me if it's more than $500."*

The LLM will pause, call the appropriate NestJS tool, read the database results, and formulate a natural language response.

## 📂 Project Structure
*src/user/, src/product/, src/order/ - NestJS modules containing Controllers, Services, Models, and MCP Resolvers.*

**src/chat.ts** - The Vercel AI SDK client implementing the conversational loop.

**src/seed.ts** - Database seeding script.