Skip to content
Agentic Levels
Start hereLevelsTracksPromptsAbout
Get my levelGet my level

  • Start hereYour first useful task with AI.
  • Level quiz8 questions. No AI needed.
  • Fluency ScoreYour Level and a score, from your own AI.

  • LevelsThe 11 levels, in order.
  • LessonsShort lessons, filter by level.
  • TracksGuided paths by tool, role or goal.
  • Set up your AI tool
  • Starter packFree files that help AI remember.

  • PromptsReady prompts to copy and adapt.
  • ChatGPT vs Claude vs GeminiPick the right one for the job.
  • How-tos
  • ReferenceCommands for Claude Code, Codex, Antigravity.
  • ResourcesHand-picked reading and tools.

  • Your progress
  • Your next moveAn interview that picks your next skill.
  • Your next lessonsThe next lessons at your level.
  • Updates

  • Pricing
  • About
  • FAQ & Feedback
  • Work with us
  • How we handle your data
Sign inPreferences
AboutPricingHow-tosPromptsStarter packMy stackHow we handle your data

© 2026 Fuentes Studio

PrivacyTermsRefundsAccessibilityAI content policyCredits
yourCouncil
Ready to help
✦

What do you want to understand?

Ask anything about what you're learning.

←How-to guides
  1. Resources
  2. ›How-to guides

How to use Microsoft Agent Framework

Microsoft's enterprise-ready multi-agent framework: stable APIs, long-term support, multi-provider models, and cross-runtime orchestration via A2A and MCP.

Levels 5 to 7On GitHub

AutoGen is now in maintenance mode (bug fixes only). For new multi-agent projects, Microsoft's Agent Framework is the supported successor with stable versioned APIs, long-term support commitments, and first-class support for Claude, GPT, and Gemini in the same workflow. It handles cross-runtime agent coordination via A2A protocol and integrates MCP servers natively, making it the right foundation for production multi-agent systems you plan to maintain.

Before you start

Python 3.10+ • pip install microsoft-agent-framework • API keys for at least one model provider (Anthropic, OpenAI, or Google). If you have existing AutoGen code, the framework provides migration guides.

  1. 1Understand the two-tier architecture
  2. 2Define agents with stable typed APIs
  3. 3Connect agents via A2A protocol
  4. 4Add MCP server integration

Steps

01

Understand the two-tier architecture

Agent Framework separates agents (stateless reasoning units) from runtimes (execution environments that manage message passing and lifecycle). This split lets you run the same agent logic locally during development and on Azure or another cloud runtime in production without changing agent code.

Design your agent topology
I want to build a multi-agent system using Microsoft Agent Framework that does [MULTI_AGENT_TASK]. Walk me through: (1) how many agents I need and what each one does, (2) which runtime to use for development vs production, (3) how agents communicate, (4) where human oversight checkpoints belong. Give me a minimal working example.
[MULTI_AGENT_TASK]Your multi-agent workflow e.g. code generation with automated review and test execution, three agents coordinating before final output
02

Define agents with stable typed APIs

Agent Framework uses typed message schemas, agents declare what messages they send and receive. This makes agent boundaries explicit and prevents the loose string-passing that makes AutoGen workflows hard to debug at scale.

Scaffold a typed agent
Using Microsoft Agent Framework, create an agent called [AGENT_NAME] that: receives [INPUT_SCHEMA], uses [MODEL_PROVIDER] as its LLM, calls these tools: [TOOLS], and produces [OUTPUT_SCHEMA]. Include full type annotations and a docstring.
[AGENT_NAME]Your agent name e.g. CodeReviewAgent
[INPUT_SCHEMA]What this agent receives e.g. a PR diff as a string
[MODEL_PROVIDER]LLM provider e.g. Anthropic Claude Sonnet
[TOOLS]Tools available to this agent e.g. read_file, search_codebase, post_review_comment
[OUTPUT_SCHEMA]What this agent produces e.g. a ReviewResult with severity, findings list, and approve/reject decision
03

Connect agents via A2A protocol

A2A (Agent-to-Agent) lets agents running in different runtimes, or built with different frameworks, communicate over a standard protocol. Wire your agents together by declaring their A2A endpoints, then test communication with a simple end-to-end task before adding complexity.

Wire multi-agent communication
I have these agents in my Agent Framework setup: [AGENT_LIST]. Define the A2A message routing between them: which agent initiates, how results flow, and how the workflow terminates. Show the runtime configuration for local development.
[AGENT_LIST]Your agents e.g. OrchestratorAgent, CoderAgent, ReviewerAgent, TestRunnerAgent
04

Add MCP server integration

Agent Framework natively exposes agents as MCP servers and can consume external MCP tools. This means your agents can be called from Claude Code, Cursor, or any MCP client without additional wiring.

Expose an agent as an MCP server
Show me how to expose my [AGENT_NAME] from Microsoft Agent Framework as an MCP server that Claude Code can call. Include the server configuration and how to add it to a Claude Code settings.json.
[AGENT_NAME]Agent to expose e.g. CodeReviewAgent

How do you know it worked?

  • Two agents exchange typed messages end-to-end without manual turn management.
  • You can swap the LLM provider (e.g. Claude → GPT) by changing one config line.
  • A2A communication works between at least two agents in separate runtime instances.
  • At least one agent is reachable as an MCP tool from Claude Code.
  • Human-in-the-loop approval checkpoint fires before any irreversible action.

If something looks off

Do not start new projects on AutoGen
Microsoft AutoGen is in maintenance mode, security patches and bug fixes only. New multi-agent projects should use Agent Framework. Migration guides exist if you have existing AutoGen code.
Define termination conditions explicitly
Multi-agent loops can run indefinitely without explicit stopping criteria. Every workflow must declare when it terminates, by task completion, turn count, or explicit human approval.
Profile cost before scaling agent count
Cost scales with agents × turns × tokens per turn. Benchmark a two-agent workflow first, then extrapolate before deploying a larger topology to production.

Skills & Commands Reference

Core Agent Framework concepts and API patterns for building production multi-agent systems.

Command / SkillWhat it doesWhen to use it
Agent classStateless reasoning unit, declares input/output schemas and uses a configured LLMBase building block for every workflow, one agent per distinct role or capability
RuntimeExecution environment managing agent lifecycle and message passing, local for dev, Azure/cloud for productionChoose local runtime during development; swap to cloud runtime without changing agent code
A2A protocoladvancedAgent-to-Agent communication standard enabling cross-runtime and cross-framework agent coordinationConnecting agents in different runtimes, or integrating with agents built on other frameworks
MCP integrationadvancedExpose any agent as an MCP server, or consume external MCP tools from within agentsMaking agents callable from Claude Code, Cursor, or any MCP-compatible client
Multi-provider model configConfigure Claude, GPT, Gemini, or any provider per agent, swap via config, not codeRouting different agents to different models by cost, capability, or latency requirements
Human-in-the-loop checkpointsPause execution and request human approval before proceeding to next agent or irreversible actionAny workflow step that deploys, sends, or permanently modifies data

See also

  • LangGraph: stateful agent graphs with durable execution
  • Ruflo: enterprise swarm platform

What you'll do, 4 steps

  1. 1Understand the two-tier architecture
  2. 2Define agents with stable typed APIs
  3. 3Connect agents via A2A protocol
  4. 4Add MCP server integration

Was this helpful?