Skip to main content
The AI Toolkit extension for Visual Studio Code streamlines agent development. It includes an Agent (Prompt) Builder that lets you wire up MCP servers to a language model and test tool invocation with natural-language prompts — all without leaving the editor.
The AI Toolkit currently supports Python and TypeScript MCP servers.

​
Learning objectives

By the end of this lesson you will be able to:
  • Consume an MCP server from the AI Toolkit Agent Builder
  • Configure an agent with a system prompt and model
  • Use MCP tools via natural-language prompts
  • Add new tools to an MCP server and reload the agent

​
Prerequisites

Visual Studio Code

Download and install VS Code if you haven’t already.

AI Toolkit extension

Install AI Toolkit from the VS Code Marketplace.
If you are on macOS, there is a known issue with dependency installation. Check the AI Toolkit docs for the latest status before starting this exercise.

​
Exercise: Building a calculator agent

1

Add the GPT-4o model to AI Toolkit

  1. Open the AI Toolkit extension from the Activity Bar.
  2. In the Catalog section, select Models to open the Model Catalog.
  3. Search for OpenAI GPT-4o and click + Add — make sure to select the GitHub-hosted variant.
  4. Confirm the model appears in My Models in the Activity Bar.
2

Create an agent

  1. In the AI Toolkit Tools section, select Agent (Prompt) Builder.
  2. Click + New Agent and enter the name Calculator Agent.
  3. For the Model field, select OpenAI GPT-4o (via GitHub).
3

Generate a system prompt

  1. Click the Generate system prompt button.
  2. In the Generate a prompt window, enter: 
    You are a helpful and efficient math assistant. When given a problem
involving basic arithmetic, you respond with the correct result.
  3. Click Generate. The system prompt will appear in the System prompt field.
  4. Review and modify the prompt if needed.
4

Create the calculator MCP server

  1. In the Tools section of the Agent Builder, click + MCP Server.
  2. Select + Add Server → Create a New MCP Server.
  3. Choose the python-weather template.
  4. Select Default folder as the save location.
  5. Name the server Calculator.
  6. A new VS Code window opens. Select Yes, I trust the authors.
  7. In the terminal, create and activate a virtual environment: 
    python -m venv .venv
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate
  8. Install the dependencies: 
    pip install -e .[dev]
  9. Open src/server.py and replace its contents with the calculator implementation: 
    """
Sample MCP Calculator Server implementation in Python.
Provides basic arithmetic operations: add, subtract, multiply, divide.
"""
from mcp.server.fastmcp import FastMCP

server = FastMCP("calculator")

@server.tool()
def add(a: float, b: float) -> float:
    """Add two numbers together and return the result."""
    return a + b

@server.tool()
def subtract(a: float, b: float) -> float:
    """Subtract b from a and return the result."""
    return a - b

@server.tool()
def multiply(a: float, b: float) -> float:
    """Multiply two numbers together and return the result."""
    return a * b

@server.tool()
def divide(a: float, b: float) -> float:
    """Divide a by b. Raises ValueError if b is zero."""
    if b == 0:
        raise ValueError("Cannot divide by zero")
    return a / b
5

Run the agent

  1. Press F5 to start debugging the MCP server. The Agent Builder opens.
  2. In the User prompt field, enter: 
    I bought 3 items priced at $25 each, and then used a $20 discount.
How much did I pay?
  3. Click Run. The agent should respond that you paid $55.
  4. Inspect the Tool Response panel to see the individual tool calls:
    • multiply(3, 25) → 75
    • subtract(75, 20) → 55
  5. Try more prompts to exercise additional tools.
  6. When finished, stop the server with Ctrl/Cmd+C in the terminal.

​
Assignment

Add a square_root tool to server.py: 
import math

@server.tool()
def square_root(n: float) -> float:
    """Return the square root of n. Raises ValueError for negative numbers."""
    if n < 0:
        raise ValueError("Cannot take square root of a negative number")
    return math.sqrt(n)
Restart the server (F5) and submit prompts that exercise the new tool, such as:
  • “What is the square root of 225?”
  • “I need the square root of 16 plus 7.”

​
Key takeaways

  • The AI Toolkit Agent Builder is a complete environment for building and testing MCP-powered agents.
  • You can add new tools to MCP servers at any time and reload the agent to pick them up.
  • The AI Toolkit includes Python and TypeScript templates to accelerate server creation.
  • Natural-language prompts trigger tool calls automatically — the LLM selects and invokes the right tools.

Build docs developers (and LLMs) love

Get started for freeTalk to us