Microsoft Agent Framework

AI Agents

Microsoft

FreeOpen Source

About

Microsoft's open-source agent framework, the official successor to AutoGen and Semantic Kernel: multi-agent workflows in .NET/Python/Go with native MCP and A2A support

Visit

Share this tool

Twitter LinkedIn Facebook

Our Verdict

Recommended

The default agent framework for Microsoft-stack enterprises

Microsoft did the right thing: instead of maintaining two overlapping projects, it merged AutoGen's multi-agent research DNA with Semantic Kernel's enterprise rigor into one supported framework. For .NET shops and Azure-centric organizations, this is now the obvious path — open standards, official migration guides and a long support runway. Outside the Microsoft orbit, LangGraph's ecosystem is still deeper, but nobody building on Azure should look elsewhere first.

Best for

•Enterprise .NET and Azure-centric engineering teams
•Teams migrating off Semantic Kernel or AutoGen
•Organizations standardizing on MCP and A2A interoperability

Consider alternatives if

•You want the deepest Python ecosystem (→ LangGraph)
•You need type-safe Python with minimal abstraction (→ Pydantic AI)

Try Microsoft Agent Framework

Supported Platforms

WindowsmacOSLinux

Available platforms include Windows, macOS, and Linux.

Key Features

Official successor to AutoGen and Semantic Kernel, merged into one framework

Multi-agent workflows and orchestration in .NET, Python and Go

Native support for open standards: Model Context Protocol (MCP) and Agent-to-Agent (A2A)

Works with any model provider — OpenAI, Azure OpenAI, Anthropic and more

Deep Azure AI Foundry integration for hosted agents, observability and compliance

Official migration guides from Semantic Kernel and AutoGen projects

Pricing

free

Fully free and open source (MIT license): the framework costs nothing, you pay only your model provider or Azure usage

paid

No paid tiers for the framework itself; hosted agents and enterprise features on Azure AI Foundry are billed as standard Azure consumption

Use Cases

Enterprise .NET shops adding production agents to existing systems

Migrating Semantic Kernel or AutoGen projects to a supported successor

Multi-agent workflows deployed and governed on Azure

Interoperable agents built on MCP and A2A open standards

Pros

One official framework instead of two overlapping Microsoft projects

Rare multi-language parity across .NET, Python and Go

Open standards first: MCP and A2A baked in from day one

Enterprise-grade backing, security posture and Azure integration

Cons

1.0 shipped in April 2026 — the ecosystem is still young

Migrating from Semantic Kernel/AutoGen requires real code changes

The best experience assumes an Azure-centric stack

Community content and third-party integrations trail LangChain's

Latest Update

April 3, 2026: version 1.0 GA officially merges Semantic Kernel and AutoGen into one production SDK; BUILD 2026 adds Agent Harness, Hosted Agents and CodeAct

More from Microsoft

Phi-4

Microsoft

Microsoft's small yet powerful LLM, 14B parameters with performance rivaling much larger models, deployable locally and on cloud

Microsoft Copilot Studio

Microsoft

Microsoft's low-code AI agent building platform: create agents graphically or with natural language, publish to Teams/websites/M365, deeply integrated with Power Platform