Get Started Beginner ⏱️ 8 min

Introduction

Welcome to the bootcamp. You'll go from your first agent to a secured, observable, multi-agent platform — all with the Microsoft Agent Framework for Python (1.8.x).

🏢
Who this is for
Engineers and architects shipping AI agents into real products. We favour production concerns — auth, tools, threads, observability — over toy demos.

The mental model

Everything in the framework composes from a few building blocks. A chat client talks to a model; an agent wraps that client with instructions, tools, and memory; workflows and multi-agent patterns coordinate many agents; and MCP lets agents reach tools across a network boundary.

Agent Framework building blocks
flowchart TD
    CC[Chat Client] --> A[Agent]
    A --> FT[Function Tools]
    A --> HT[Hosted & MCP Tools]
    A --> TH[Threads]
    A --> WF[Workflows]
    WF --> MA[Multi-Agent]
    MA --> A2A[Agent-to-Agent]
    MCP[MCP] --> RM[Remote MCP Servers]
    RM --> HT
    classDef hi fill:#6366f1,stroke:#4f46e5,color:#fff;
    class A hi;
From a single chat client to coordinated multi-agent systems over MCP.

What you'll build

🚀
Get Started
Environment, first agent, providers.
🧱
Core Foundations
Tools, threads, streaming.
🔀
Tools & Workflows
Structured output and orchestration.
🛰️
Multi-Agent
Supervisors, A2A, remote MCP.

What you need

  1. Python 3.13+ — the bootcamp targets the current runtime.
  2. uv — fast environment and dependency management.
  3. Azure OpenAI or Microsoft Foundry — a deployed chat model.
  4. Comfort with async/await — agent runs are asynchronous.
💡
Read it like docs, build it like a lab
Every lesson pairs a concept with a runnable modules/ folder. Read, then run.