Version 1.0 of Microsoft Agent Framework, which is used to build and execute agents as well as complex actions across multiple AI agents, is now available in production for both .NET and Python.
The framework provides tools to build multi-agent orchestration, supports models from multiple providers, and enables interoperability between agents through its support for Agent2Agent Protocol (A2A) and Model Context Protocol (MCP). In a blog post announcing the release, the company said that getting started with the framework takes just a few lines of code.
Development and AI agent experts I spoke to in recent weeks cited a use case example in which the framework can be used to orchestrate multiple agents for sales lead data prioritization and a related email campaign. They noted the system’s graph-based workflows simplify orchestration. They also highlighted the value in being able to choose model providers and have a range of agents and back-end systems interoperate.
Core Framework Functions
The Version 1.0 release incorporates a wide range of functions, and Microsoft also outlined features in the framework pipeline. I’m presenting highlights of both current and planned features below.
Current release features include:
- A graph-based engine that developers can use to build workflows combining agent reasoning with business logic, conditional branches, and converge results, all while surviving interruptions during long-running processes.
- Support for orchestration patterns including sequential, concurrent, handoff, group chat, and Microsoft’s Magentic-One multi-agent system. All patterns support streaming, checkpointing, human-in-the-loop approvals, and pause/resume capability.
- Ability to define agents’ instructions, tools, memory configuration, and orchestration topology, then to load and run them with a single API call.
- MCP support so agents can dynamically discover and invoke external tools exposed over MCP-compliant servers. A2A support that enables cross-runtime agent collaboration so agents can coordinate with agents running in other frameworks.
- First-party service connectors for agent interactions with Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic Claude, Amazon Bedrock, and Google Gemini.
- Middleware hooks to intercept, transform, and extend agent behavior at execution with content safety filters, logging, compliance policies, and custom logic – all without modifying agent prompts.
Future Features
Version 1.0 is also shipping with preview features that are available for early adoption; Microsoft noted these features could evolve based on community feedback. They currently include:
- A developer interface with a local debugger for visualizing agent execution, message flows, tool calls, and orchestration decisions.
- GitHub Copilot SDK and Claude Code SDK to utilize either platform as an agent harness directly from within Agent Framework’s orchestration code. The SDKs handle the autonomous agent loop – planning, tool execution, file edits, and session management – while Agent Framework wraps them into a multi-agent workflow.
- A customizable Microsoft Agent Framework harness and local runtime giving agents access to shell, file system, and messaging for coding agents, automation, and personal assistant functions.
- Integration into Microsoft Foundry to operate Agent Framework agents as managed services on Microsoft Foundry or as Azure Durable Functions
- Adapters for CopilotKit and ChatKit, including tool execution status and human-in-the-loop flows, for agent output to CopilotKit and ChatKit
- Integration with Foundry’s managed tools and memory as well as observability and evaluations dashboards
- Reusable domain “skills” that give agents structured capabilities out of the box
In terms of next steps, Microsoft recommends the Agent Framework quickstart guide for walkthroughs in .NET and Python or accessing Agent Framework samples on GitHub.
Ask Cloud Wars AI Agent about this analysis
