WORKSHOP GUIDE - SESSION 3

AI Agents & MCP

Building Agents that can Use Tools, APIs, and Reason.

Part 1: Key Concepts

Agents vs Chatbots

A Chatbot reads and writes. An Agent can Act. It has "Tools" (like Python functions) that it can call to get real-time data or perform tasks.

MCP (Model Context Protocol)

The standard way to connect LLMs to external data. Instead of hardcoding API calls, we run a small local server (MCP Server) that exposes tools to the AI in a safe, standardized way.

Part 2: The Lab - "The Finance Analyst"

The Mission

You are a Quantitative Analyst. You need to build a custom AI agent that can fetch LIVE stock data using the YFinance API and give you immediate investment advice.

Prerequisites

Ensure your Python MCP Server is running:

python mcp_server.py

Part 3: The Prompt Library

Level 1: Basic Tool Call

Goal: Verify the Agent can call the tool.

What is the current stock price of Apple (AAPL)?
Watch: Look at the "Thought Chain". You should see `Action: get_stock_price("AAPL")`.

Level 2: Comparison (Multi-Step)

Goal: Force the agent to make multiple calls.

Compare the price of Apple (AAPL) and Microsoft (MSFT). Which one is more expensive per share right now?

Level 3: Analysis & Synthesis

Goal: Combine Data with Reasoning.

Fetch the last 5 days of data for Tesla (TSLA). Is the trend bullish or bearish? Explain your reasoning based on the moving average.

Level 4: Handling Errors (Robustness)

Goal: Test error recovery.

Check the price of a fake company called 'XYZABC'. If it fails, check Google (GOOGL) instead.
Insight: The Agent should receive an error from the first call, catch it, and proceed to the second instruction autonomously.