Welcome! In this tutorial, we will build a simple AI-powered backend application that integrates an AI model with tools and data using the Model Context Protocol (MCP) — all on your local machine.
No experience with AI or coding? No problem! This guide will walk you through every step, like a friendly tutor.
🧠What is Model Context Protocol (MCP)?
Imagine you’re talking to a super-smart assistant (like ChatGPT). Now imagine that assistant can look things up, calculate results, and read data from your files — just like a helpful human would.
MCP is the set of rules that tells the AI:
"Hey, if you see a question that needs a calculator, use it!"
"If the user asks about data, check the file first before guessing."
Think of it as a walkie-talkie between the AI and real tools/data.
🛠️ Step 1: Set Up Your Environment
1.1 Install Python
If you don’t already have Python:
Download and install Python 3.9 or higher.
Make sure to check "Add Python to PATH" during installation.
To check if Python is installed:
python --version1.2 Create a Folder for Your Project
mkdir mcp-ai-demo
cd mcp-ai-demo1.3 Set Up a Virtual Environment (Optional but Recommended)
python -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate1.4 Install Required Packages
pip install openai pandas📦 Step 2: Prepare Your Data (CSV File)
We’ll use a simple product catalog as our "database."
Create a file called products.csv:
id,name,price,stock
1,Red T-shirt,19.99,25
2,Blue Jeans,49.99,10
3,Green Hat,14.99,5
4,Black Sneakers,89.99,2🧠Step 3: Set Up AI Access (GPT Model)
3.1 Get an OpenAI API Key
Create a free account and get your key.
3.2 Store the Key in Your Code (DO NOT share it)
We’ll use a .env-style setup for this tutorial.
Create a file config.py:
# config.py
OPENAI_API_KEY = "your-api-key-here" # Replace this with your real key🧰 Step 4: Create Tools the AI Can Use
4.1 Data Tool (Search CSV File)
# data_tool.py
import pandas as pd
def search_products(keyword):
df = pd.read_csv('products.csv')
results = df[df['name'].str.contains(keyword, case=False)]
return results.to_dict(orient='records')4.2 Calculator Tool
# calculator_tool.py
def calculate(expression):
try:
return eval(expression)
except Exception as e:
return str(e)🧠Step 5: Connect AI via MCP Logic
5.1 Main Script
# main.py
import openai
from config import OPENAI_API_KEY
from data_tool import search_products
from calculator_tool import calculate
openai.api_key = OPENAI_API_KEY
def ai_call(user_input):
print("\nUser asked:", user_input)
# Step 1: Let AI decide what to do
prompt = f"""
You are a smart AI assistant. You can:
1. Search product data using 'search_products(keyword)'
2. Use a calculator with 'calculate(expression)'
If the user asks something like "find red shirt" -> call search_products('red shirt')
If the user says "what's 5 * 20" -> call calculate('5 * 20')
Otherwise, reply directly.
Now, here's what the user asked:
"{user_input}"
What should you do?
"""
response = openai.Completion.create(
engine="text-davinci-003",
prompt=prompt,
max_tokens=150,
temperature=0
)
action = response.choices[0].text.strip()
print("\nAI decided:", action)
# Step 2: Execute based on AI decision
if "search_products(" in action:
keyword = action.split("(")[1].split(")")[0].strip("'\"")
result = search_products(keyword)
print("\nSearch Results:", result)
elif "calculate(" in action:
expr = action.split("(")[1].split(")")[0].strip("'\"")
result = calculate(expr)
print("\nCalculation Result:", result)
else:
print("\nAI Response:", action)
# Simple CLI Loop
if __name__ == "__main__":
while True:
user_input = input("\nAsk something (or type 'exit'): ")
if user_input.lower() == 'exit':
break
ai_call(user_input)▶️ Step 6: Run and Test It Locally
In your terminal:
python main.pyTry typing:
Find red t-shirtHow much is 15 * 3.5?Show green hat
🎉 You’ve just built a mini AI system that talks to tools and data!
💡 What to Try Next
Add a second data file (like customer info).
Add a weather tool using a public API.
Let the AI update your CSV (e.g., mark item as "out of stock").
Try switching the AI model to something open-source (e.g., using HuggingFace Transformers).
📦 Summary
| Component | Description |
|---|---|
| CSV File | Acts as your mock database |
| AI Model | Interprets user queries |
| Tools | Execute actions (search, calculate, etc.) |
| MCP Logic | Bridges AI intent with tool invocation |
You now understand the basics of Model Context Protocol — using plain Python and local files. 🧠🔌📊
Happy hacking!
