Get Started with Microsoft Agent Framework

Highlights

• Flexible Agent Framework: build, orchestrate, and deploy AI agents and multi-agent systems
• Multi-Agent Orchestration: Group chat, sequential, concurrent, and handoff patterns
• Plugin Ecosystem: Extend with native functions, OpenAPI, Model Context Protocol (MCP), and more
• LLM Support: OpenAI, Foundry, Anthropic, and more
• Runtime Support: In-process and distributed agent execution
• Multimodal: Text, vision, and function calling
• Cross-Platform: .NET and Python implementations

Quick Install

pip install agent-framework-core
# Optional: Add Microsoft Foundry integration
pip install agent-framework-foundry
# Optional: Add OpenAI integration
pip install agent-framework-openai

Supported Platforms:

• Python: 3.11–3.15
• OS: Windows, macOS, Linux

Python 3.10 is no longer supported because it has reached upstream end of life. Package metadata still permits Python 3.10 installation, but CI no longer runs unit or integration tests on it and compatibility is not guaranteed. Use Python 3.11–3.15 for supported environments, subject to integration-specific limits. Raising the installation minimum will be handled separately.

1. Setup API Keys

Depending on the client you want to use, there are various environment variables you can set to configure the chat clients. This can be done in the environment itself, or with a .env file in your project root, some examples of environment variables include:

FOUNDRY_PROJECT_ENDPOINT=...
FOUNDRY_MODEL=...
...
OPENAI_API_KEY=sk-...
OPENAI_CHAT_COMPLETION_MODEL=...
OPENAI_CHAT_MODEL=...
...
AZURE_OPENAI_API_KEY=...
AZURE_OPENAI_ENDPOINT=...
AZURE_OPENAI_MODEL=...

You can also override environment variables by explicitly passing configuration parameters to the chat client constructor:

from agent_framework.openai import OpenAIChatClient

client = OpenAIChatClient(
    api_key="",
    model="",
)

Telemetry controls

Agent Framework adds its package/version User-Agent to supported client requests. Approved Microsoft Foundry and Azure OpenAI request paths can also carry a documented feature-usage token.

• AGENT_FRAMEWORK_FEATURE_MASK_DISABLED=true disables only the feature-usage token while retaining the package/version User-Agent.
• AGENT_FRAMEWORK_USER_AGENT_DISABLED=true disables the entire Agent Framework User-Agent contribution, including the feature token.

OpenTelemetry providers configured through configure_otel_providers() share a resource containing "microsoft.agent-framework": "True" (a string) by default. This resource attribute applies to logs, traces, and metrics, including other instrumentation using those providers. Existing environment and explicit resource attribute overrides retain their precedence. If you configure providers yourself, use agent_framework.observability.create_resource() for each provider or add the attribute to your own shared resource.

See the following getting started samples for more information.

2. Create a Simple Agent

Create agents and invoke them directly:

import asyncio
from agent_framework import Agent
from agent_framework.openai import OpenAIChatClient

agent = Agent(
    client=OpenAIChatClient(),
    instructions="""
    1) A robot may not injure a human being...
    2) A robot must obey orders given it by human beings...
    3) A robot must protect its own existence...

    Give me the TLDR in exactly 5 words.
    """
)

result = asyncio.run(agent.run("Summarize the Three Laws of Robotics"))
print(result)
# Output: Protect humans, obey, self-preserve, prioritized.

3. Directly Use Chat Clients (No Agent Required)

You can use the chat client classes directly for advanced workflows:

import asyncio
from agent_framework.openai import OpenAIChatClient
from agent_framework import Message, Role

async def main():
    client = OpenAIChatClient()

    response = await client.get_response([
        Message("system", ["You are a helpful assistant."]),
        Message("user", ["Write a haiku about Agent Framework."])
    ])
    print(response.messages[0].text)

    """
    Output:

    Agents work in sync,
    Framework threads through each task—
    Code sparks collaboration.
    """

asyncio.run(main())

4. Build an Agent with Tools and Functions

Enhance your agent with custom tools and function calling:

import asyncio
from typing import Annotated
from random import randint
from agent_framework import Agent
from agent_framework.openai import OpenAIChatClient


def get_weather(
    location: Annotated[str, "The location to get the weather for."],
) -> str:
    """Get the weather for a given location."""
    conditions = ["sunny", "cloudy", "rainy", "stormy"]
    return f"The weather in {location} is {conditions[randint(0, 3)]} with a high of {randint(10, 30)}°C."


def get_menu_specials() -> str:
    """Get today's menu specials."""
    return """
    Special Soup: Clam Chowder
    Special Salad: Cobb Salad
    Special Drink: Chai Tea
    """


async def main():
    agent = Agent(
        client=OpenAIChatClient(),
        instructions="You are a helpful assistant that can provide weather and restaurant information.",
        tools=[get_weather, get_menu_specials]
    )

    response = await agent.run("What's the weather in Amsterdam and what are today's specials?")
    print(response)

    # Output:
    # The weather in Amsterdam is sunny with a high of 22°C. Today's specials include
    # Clam Chowder soup, Cobb Salad, and Chai Tea as the special drink.

asyncio.run(main())

You can explore additional agent samples here.

MCP HTTP client ownership and cookies

Breaking change: HTTP clients created by MCPStreamableHTTPTool no longer persist response cookies. This applies with or without a header_provider, including repeated calls within the same MCP session. Connection reuse, timeout and redirect defaults, and cleanup on close or reset are preserved. Explicit Cookie headers supplied through static_headers or a header_provider are still supported.

Applications relying on cookie-based authentication, sessions, or load-balancer affinity must supply an httpx.AsyncClient through http_client=. Supplied clients retain their configuration and response-cookie handling and must be closed by the caller. Scope cookie-bearing clients and MCP tool sessions to one authenticated principal; do not share a user-specific cookie jar across users. Rejecting cookies does not partition MCP protocol sessions or other server-side state between principals.

5. Multi-Agent Orchestration

Coordinate multiple agents to collaborate on complex tasks using orchestration patterns:

import asyncio
from agent_framework import Agent
from agent_framework.openai import OpenAIChatClient


async def main():
    # Create specialized agents
    writer = Agent(
        client=OpenAIChatClient(),
        name="Writer",
        instructions="You are a creative content writer. Generate and refine slogans based on feedback."
    )

    reviewer = Agent(
        client=OpenAIChatClient(),
        name="Reviewer",
        instructions="You are a critical reviewer. Provide detailed feedback on proposed slogans."
    )

    # Sequential workflow: Writer creates, Reviewer provides feedback
    task = "Create a slogan for a new electric SUV that is affordable and fun to drive."

    # Step 1: Writer creates initial slogan
    initial_result = await writer.run(task)
    print(f"Writer: {initial_result}")

    # Step 2: Reviewer provides feedback
    feedback_request = f"Please review this slogan: {initial_result}"
    feedback = await reviewer.run(feedback_request)
    print(f"Reviewer: {feedback}")

    # Step 3: Writer refines based on feedback
    refinement_request = f"Please refine this slogan based on the feedback: {initial_result}\nFeedback: {feedback}"
    final_result = await writer.run(refinement_request)
    print(f"Final Slogan: {final_result}")

    # Example Output:
    # Writer: "Charge Forward: Affordable Adventure Awaits!"
    # Reviewer: "Good energy, but 'Charge Forward' is overused in EV marketing..."
    # Final Slogan: "Power Up Your Adventure: Premium Feel, Smart Price!"

if __name__ == "__main__":
    asyncio.run(main())

Note: Sequential, Concurrent, Group Chat, Handoff, and Magentic orchestrations are available. See examples in orchestration samples.

More Examples & Samples

• Getting Started with Agents: Basic agent creation and tool usage
• Chat Client Examples: Direct chat client usage patterns
• Foundry Integration: Foundry integration
• Workflows Samples: Advanced multi-agent patterns

Agent Framework Documentation

• Agent Framework Repository
• Python Package Documentation
• .NET Package Documentation
• Design Documents
• Learn Documentation