MCP 服务器

ShopBot AI MCP Server

AI customer support MCP server with order status lookup and RAG-powered knowledge base search for e-commerce stores.

README

ShopBot AI — MCP-Powered Customer Support Agent

A domain-specific AI customer support agent for e-commerce, built with Python and Flask. ShopBot connects a large language model directly to a live MySQL database and a RAG knowledge base via a custom MCP server — giving it real-time access to order data and product information without hallucinating.

🟢 Live Demo: https://shopbot-ai-os4z.onrender.com

⚠️ Note: The app is hosted on Render's free tier and may take 30–60 seconds to load on first visit (cold start). Please be patient. If the bot does not reply to your message, the Gemini API free quota (20 requests/day) has been reached and will reset the following day.

The Problem It Solves

Problem: E-commerce stores receive hundreds of repetitive support requests — "Where is my order?", "What is your return policy?", "Do you have this in stock?" — that consume staff time and slow down response times.

Solution: ShopBot gives businesses a deployable AI support agent that connects directly to their order database and knowledge base. Customers get instant, accurate answers without a human in the loop.

Result: Demonstrates how small e-commerce businesses can reduce support workload, eliminate repetitive queries, and provide 24/7 accurate order information — with a fully custom stack they own and control.

What It Does

ShopBot handles two types of customer queries intelligently:

Order tracking — looks up real order status, tracking numbers, and shipping dates from a live MySQL database
Product & policy questions — searches a RAG knowledge base covering FAQs, return policy, shipping info, and product catalog

The LLM classifies each incoming message and routes it to the right tool automatically. If neither tool is needed, it answers directly.

Architecture

Browser → Flask API → Gemini 2.5 Flash (classifier)
                    ↓
              MCP Server
              ├── Tool 1: check_order_status   → MySQL Database
              └── Tool 2: search_knowledge_base → FAISS Vector Store (RAG)

MCP Tools

Tool	Description
`check_order_status`	Queries MySQL for order status, tracking number, and shipping date by order ID or customer email
`search_knowledge_base`	Performs semantic search over the FAISS vector store to answer product, policy, and FAQ questions

Tech Stack

Layer	Technology
LLM	Gemini 2.5 Flash (google-genai SDK)
MCP Server	FastMCP (stdio transport)
RAG Pipeline	LangChain + FAISS + gemini-embedding-001
Backend	Flask (async orchestration)
Database	MySQL
Frontend	HTML / CSS / JavaScript
Deployment	Render.com

Project Structure

shopbot-ai/
├── rag.py                        ← RAG pipeline (FAISS index loader)
├── mcp/shopbot_server.py         ← MCP server with 2 tools
├── flask_app/
│   ├── app.py                    ← Flask API + MCP client + Gemini calls
│   └── templates/index.html      ← Chat UI
├── knowledge_base/               ← faq.txt, return_policy.txt, shipping_info.txt, products.txt
├── faiss_db/                     ← FAISS vector index
└── Procfile                      ← Render deployment config

Local Setup

Prerequisites

Python 3.10+
MySQL running locally
Google API key (Gemini)

Installation

git clone https://github.com/hayamot0/shopbot-ai.git
cd shopbot-ai
pip install -r requirements.txt

Environment Variables

Create a .env file in the root directory:

GOOGLE_API_KEY=your_google_api_key_here
DB_HOST=localhost
DB_USER=your_mysql_user
DB_PASSWORD=your_mysql_password
DB_NAME=shopbot

Database Setup

CREATE DATABASE shopbot;

CREATE TABLE customers (
    customer_id INT PRIMARY KEY AUTO_INCREMENT,
    name VARCHAR(100),
    email VARCHAR(100),
    phone VARCHAR(20)
);

CREATE TABLE orders (
    order_id INT PRIMARY KEY AUTO_INCREMENT,
    customer_id INT,
    product VARCHAR(100),
    status VARCHAR(50),
    order_date DATE,
    tracking_number VARCHAR(50),
    FOREIGN KEY (customer_id) REFERENCES customers(customer_id)
);

Run

python flask_app/app.py

Visit http://localhost:5000

Key Design Decisions

stdio transport over SSE — chosen for simplicity and portfolio demonstration; SSE would be used for production multi-client deployments
Gemini over OpenAI — used due to regional payment gateway limitations; the architecture is LLM-agnostic and can be swapped to any provider
FAISS over hosted vector DB — local vector store keeps the project self-contained with zero additional infrastructure cost
Async event loop isolation — each Flask request spawns its own event loop to avoid conflicts between Flask's sync context and the async MCP client