Understand Microsoft Agent Framework AI agents
Unit 2 of 6 · Module 13: Develop an AI agent with Microsoft Agent Framework
Before you start
Short explainers for ideas this lesson builds on. Watch these first if they're new to you.
Go deeper
Unlocks when you've watched the lesson.
Read this lesson
A pegboard of building blocks
Picture a workshop wall where every tool hangs on its own peg. You take down only what the job needs, and add more as the job grows. Agent Framework works that way: instead of gluing separate libraries together yourself, you get building blocks that fit. Model clients give you one way to talk to many AI providers. An agent session keeps the conversation's state across turns. Context providers are memory you can plug in, bringing relevant facts to the agent. Function tools are your own functions, registered with the agent, and the framework writes their schemas for you. MCP clients let the agent discover tools from MCP servers while it runs. Middleware lets you step in before and after an agent acts, to log or change what happens. And workflows arrange several agents into sequential, concurrent, group chat, or hand-off patterns.
What every agent can do
Every agent in the framework comes from the same Agent base class, so they all behave alike, whichever model provider sits underneath. Out of the box, an agent can call the functions you give it, hold a conversation over many turns, return structured output that matches a schema, stream its answer as it's written, and use tools the provider hosts, like code execution, file search, or web search, where the provider supports them. You write your application once, instead of learning each provider's quirks.
One plug, many sockets
Think of a travel adapter: one plug, and a different socket in each country. Because the interface is shared, you can switch the model service underneath and change only the client setup, not your agent's logic. Providers include Microsoft Foundry Agent Service, Azure OpenAI, OpenAI, Anthropic Claude, Amazon Bedrock, GitHub Copilot, and Ollama. There's one big difference between them: where the conversation history lives. Foundry Agent Service, and the Responses APIs from Azure OpenAI and OpenAI, keep the history on the service side. With the others, your app keeps it. That's why Foundry is the recommended provider for production: the session carries on by itself, with persistent history, MCP tool support, and Azure integration built in.
Facts from Microsoft's documentation: Microsoft Agent Framework Overview (CC BY 4.0), rewritten in Loop's own words and examples. Narration, animation, and quiz by Loop. Microsoft, Azure, and Microsoft Foundry are trademarks of Microsoft. Loop is independent and not affiliated with or endorsed by Microsoft.