On this page

🎓 Workshop Overview

A progressive 7-exercise workshop for building AI agents and multi-agent workflows with Microsoft Agent Framework in Python. From creating your first agent to designing multi-agent pipelines with streaming events — all powered by Microsoft Foundry.

How to Use This Workshop

  1. Complete the Setup — Install prerequisites, configure Dev Container, set up Azure credentials
  2. Click an exercise in the sidebar (or the curriculum table below)
  3. Open your terminal alongside this site — each exercise tells you exactly what to code
  4. Use the 📋 Walkthrough links for conceptual guidance without seeing the full solution
  5. Track your progress with the sidebar progress bar

[!TIP] Each exercise has a starter.py with # TODO: markers. Fill in the blanks, then run check.sh to validate. Reference solutions are in src/demo*.py if you get stuck.

Skill Progression

Exercise 1   Exercise 2    Exercise 3     Exercise 4      Exercise 5       Exercise 6      Exercise 7
Run Agent → Web Search → MCP Tool → Struct Output → Multi-Agent → DevUI → Capstone
    ⭐          ⭐          ⭐⭐          ⭐⭐          ⭐⭐⭐        ⭐⭐⭐       ⭐⭐⭐⭐
 create      add hosted   MCP stdio     Pydantic       workflow       visual       design your
 & run       web search   tool via      + response_    edges +        debugger     own agent
 agent       tool         npx           format         streaming      + API
Phase Exercises Scaffolding Level What you do
🟢 Foundations 1, 2 High — fill-in-the-blank Complete TODO markers in mostly-written code
🟡 Integration 3, 4 Medium — guided Implement key functions with hints and references
🟠 Orchestration 5, 6 Low — goal-oriented Build toward a described outcome with minimal starter code
🔴 Capstone 7 None — from scratch Design and implement your own agent end-to-end

Curriculum

# Title Key Concepts Difficulty Time
Setup Setup & Installation Dev Container, Azure CLI, environment variables — 15-30 min
1 Run Your First Agent FoundryChatClient, .as_agent(), run(), async lifecycle ⭐ 15-20 min
2 Web Search Tool client.get_web_search_tool, Bing grounding, tool integration ⭐ 20-30 min
3 MCP Stdio Tool Model Context Protocol, MCPStdioTool, local tool servers ⭐⭐ 20-30 min
4 Structured Output response_format, Pydantic BaseModel, output parsing ⭐⭐ 20-25 min
5 Multi-Agent Workflow Workflow orchestration, agent edges, streaming events ⭐⭐⭐ 30-40 min
6 DevUI DevUI server, entity discovery, OpenAI-compatible API ⭐⭐⭐ 25-35 min
7 Capstone Project All prior concepts — build your own agent or workflow ⭐⭐⭐⭐ 45+ min

Total: 7 exercises across 4 phases — estimated ~5 hours

What is Microsoft Agent Framework?

Microsoft Agent Framework is a Python SDK for building AI agents backed by Microsoft Foundry. It provides a structured way to create agents that can use tools, produce structured output, and orchestrate multi-agent workflows — all with streaming support and a visual debugging UI.

Key Features

  • 🤖 Agent Creation — FoundryChatClient → .as_agent() → run() lifecycle
  • 🔍 Tool Integration — Hosted tools (Web Search) and local tools (MCP stdio)
  • 📊 Structured Output — Pydantic models with response_format parameter
  • 🔄 Workflow Orchestration — Multi-agent pipelines with WorkflowBuilder and edges
  • 📺 DevUI — Visual debugger with OpenAI-compatible API endpoint
  • 🔐 Azure-native — Entra ID authentication, Foundry project integration

Quick Start

# 1. Clone and open in Dev Container / Codespaces
git clone https://github.com/shinyay/getting-started-with-agent-framework.git
cd getting-started-with-agent-framework
# Open in VS Code → "Reopen in Container"

# 2. Configure environment
cp .env.example .env
# Edit .env with your Microsoft Foundry values

# 3. Authenticate
az login --use-device-code

# 4. Verify setup — run Exercise 1's solution
python src/demo1_run_agent.py

For detailed setup, see the Setup & Installation page.

Exercise Structure

Each exercise directory (workshop/exercises/ex<N>_*/) contains:

File Purpose
README.md Objectives, concepts, step-by-step instructions
starter.py Skeleton code with # TODO: markers
check.sh Validation script to verify your solution

Solutions are in src/demo*.py. Detailed walkthroughs are available as 📋 links in the sidebar.

Requirements

  • Azure subscription with AI Foundry project access
  • Python 3.10+ (3.11+ recommended)
  • Azure CLI installed and authenticated
  • Node.js / npx (for MCP exercises)
  • VS Code with Dev Container support (recommended)

⚠️ Version note: This workshop targets Agent Framework 1.2.2 (April 29, 2026 stable). Online documentation may describe a newer version with API differences. When in doubt, trust the code in this repository.

👉 Start the Setup →