AGENT SKILL

m365-agents-py

Microsoft 365 Agents SDK for Python. Build multichannel agents for Teams/M365/Copilot Studio with aiohttp hosting, AgentApplication routing, streaming responses, and MSAL-based auth. Triggers: "Microsoft 365 Agents SDK", "microsoft_agents", "AgentApplication", "start_agent_process", "TurnContext", "Copilot Studio client", "CloudAdapter".

License 许可
MIT
所属仓库
microsoft/skills
最近更新
2026-11-14
Stars
★ 3.2k
Forks
368

01 前言 Front Matter

name
m365-agents-py
license
MIT
metadata
{ "author": "Microsoft", "package": "microsoft-agents-hosting-core, microsoft-agents-hosting-aiohttp, microsoft-agents-activity, microsoft-agents-authentication-msal, microsoft-agents-copilotstudio-client", "version": "1.0.0" }
description
Microsoft 365 Agents SDK for Python. Build multichannel agents for Teams/M365/Copilot Studio with aiohttp hosting, AgentApplication routing, streaming responses, and MSAL-based auth. Triggers: "Microsoft 365 Agents SDK", "microsoft_agents", "AgentApplication", "start_agent_process", "TurnContext", "Copilot Studio client", "CloudAdapter".

02 资源文件 29

03 SKILL.md 正文

Microsoft 365 Agents SDK (Python)

Build enterprise agents for Microsoft 365, Teams, and Copilot Studio using the Microsoft Agents SDK with aiohttp hosting, AgentApplication routing, streaming responses, and MSAL-based authentication.

Before implementation

  • Use the microsoft-docs MCP to verify the latest API signatures for AgentApplication, start_agent_process, and authentication options.
  • Confirm package versions on PyPI for the microsoft-agents-* packages you plan to use.

Important Notice - Import Changes

⚠️ Breaking Change: Recent updates have changed the Python import structure from microsoft.agents to microsoft_agents (using underscores instead of dots).

Installation

pip install microsoft-agents-hosting-core
pip install microsoft-agents-hosting-aiohttp
pip install microsoft-agents-activity
pip install microsoft-agents-authentication-msal
pip install microsoft-agents-copilotstudio-client
pip install python-dotenv aiohttp

Environment Variables (.env)

CONNECTIONS__SERVICE_CONNECTION__SETTINGS__CLIENTID=<client-id>
CONNECTIONS__SERVICE_CONNECTION__SETTINGS__CLIENTSECRET=<client-secret>
CONNECTIONS__SERVICE_CONNECTION__SETTINGS__TENANTID=<tenant-id>

# Optional: OAuth handlers for auto sign-in
AGENTAPPLICATION__USERAUTHORIZATION__HANDLERS__GRAPH__SETTINGS__AZUREBOTOAUTHCONNECTIONNAME=<connection-name>

# Optional: Azure OpenAI for streaming (AAD auth via DefaultAzureCredential)
AZURE_OPENAI_ENDPOINT=<endpoint>
AZURE_OPENAI_API_VERSION=<version>

# Optional: Copilot Studio client
COPILOTSTUDIOAGENT__ENVIRONMENTID=<environment-id>
COPILOTSTUDIOAGENT__SCHEMANAME=<schema-name>
COPILOTSTUDIOAGENT__TENANTID=<tenant-id>
COPILOTSTUDIOAGENT__AGENTAPPID=<app-id>

Authentication & Lifecycle

? Two rules apply to every code sample below:

  1. This SDK is async-first — use async def handlers and async with throughout.
  2. Use explicit auth managers and context-managed network resources. Use MsalConnectionManager for agent auth, and wrap per-request HTTP resources in async with (for example, aiohttp.ClientSession).

Snippets may abbreviate this setup, but production code should always follow both rules.

Core Workflow: aiohttp-hosted AgentApplication

import logging
from os import environ

from dotenv import load_dotenv
from aiohttp.web import Request, Response, Application, run_app

from microsoft_agents.activity import load_configuration_from_env
from microsoft_agents.hosting.core import (
    Authorization,
    AgentApplication,
    TurnState,
    TurnContext,
    MemoryStorage,
)
from microsoft_agents.hosting.aiohttp import (
    CloudAdapter,
    start_agent_process,
    jwt_authorization_middleware,
)
from microsoft_agents.authentication.msal import MsalConnectionManager

# Enable logging
ms_agents_logger = logging.getLogger("microsoft_agents")
ms_agents_logger.addHandler(logging.StreamHandler())
ms_agents_logger.setLevel(logging.INFO)

# Load configuration
load_dotenv()
agents_sdk_config = load_configuration_from_env(environ)

# Create storage and connection manager
STORAGE = MemoryStorage()
CONNECTION_MANAGER = MsalConnectionManager(**agents_sdk_config)
ADAPTER = CloudAdapter(connection_manager=CONNECTION_MANAGER)
AUTHORIZATION = Authorization(STORAGE, CONNECTION_MANAGER, **agents_sdk_config)

# Create AgentApplication
AGENT_APP = AgentApplication[TurnState](
    storage=STORAGE, adapter=ADAPTER, authorization=AUTHORIZATION, **agents_sdk_config
)


@AGENT_APP.conversation_update("membersAdded")
async def on_members_added(context: TurnContext, _state: TurnState):
    await context.send_activity("Welcome to the agent!")


@AGENT_APP.activity("message")
async def on_message(context: TurnContext, _state: TurnState):
    await context.send_activity(f"You said: {context.activity.text}")


@AGENT_APP.error
async def on_error(context: TurnContext, error: Exception):
    await context.send_activity("The agent encountered an error.")


# Server setup
async def entry_point(req: Request) -> Response:
    agent: AgentApplication = req.app["agent_app"]
    adapter: CloudAdapter = req.app["adapter"]
    return await start_agent_process(req, agent, adapter)


APP = Application(middlewares=[jwt_authorization_middleware])
APP.router.add_post("/api/messages", entry_point)
APP["agent_configuration"] = CONNECTION_MANAGER.get_default_connection_configuration()
APP["agent_app"] = AGENT_APP
APP["adapter"] = AGENT_APP.adapter

if __name__ == "__main__":
    run_app(APP, host="localhost", port=environ.get("PORT", 3978))

AgentApplication Routing

import re
from microsoft_agents.hosting.core import (
    AgentApplication, TurnState, TurnContext, MessageFactory
)
from microsoft_agents.activity import ActivityTypes

AGENT_APP = AgentApplication[TurnState](
    storage=STORAGE, adapter=ADAPTER, authorization=AUTHORIZATION, **agents_sdk_config
)

# Welcome handler
@AGENT_APP.conversation_update("membersAdded")
async def on_members_added(context: TurnContext, _state: TurnState):
    await context.send_activity("Welcome!")

# Regex-based message handler
@AGENT_APP.message(re.compile(r"^hello$", re.IGNORECASE))
async def on_hello(context: TurnContext, _state: TurnState):
    await context.send_activity("Hello!")

# Simple string message handler
@AGENT_APP.message("/status")
async def on_status(context: TurnContext, _state: TurnState):
    await context.send_activity("Status: OK")

# Auth-protected message handler
@AGENT_APP.message("/me", auth_handlers=["GRAPH"])
async def on_profile(context: TurnContext, state: TurnState):
    token_response = await AGENT_APP.auth.get_token(context, "GRAPH")
    if token_response and token_response.token:
        # Use token to call Graph API
        await context.send_activity("Profile retrieved")

# Invoke activity handler
@AGENT_APP.activity(ActivityTypes.invoke)
async def on_invoke(context: TurnContext, _state: TurnState):
    invoke_response = Activity(
        type=ActivityTypes.invoke_response, value={"status": 200}
    )
    await context.send_activity(invoke_response)

# Fallback message handler
@AGENT_APP.activity("message")
async def on_message(context: TurnContext, _state: TurnState):
    await context.send_activity(f"Echo: {context.activity.text}")

# Error handler
@AGENT_APP.error
async def on_error(context: TurnContext, error: Exception):
    await context.send_activity("An error occurred.")

Streaming Responses with Azure OpenAI

from openai import AsyncAzureOpenAI
from azure.identity import DefaultAzureCredential, get_bearer_token_provider
from microsoft_agents.activity import SensitivityUsageInfo

# AAD token provider (preferred over AZURE_OPENAI_API_KEY)
token_provider = get_bearer_token_provider(
    DefaultAzureCredential(),
    "https://cognitiveservices.azure.com/.default",
)

# Module-level singleton: client lives for the agent app lifetime.
CLIENT = AsyncAzureOpenAI(
    api_version=environ["AZURE_OPENAI_API_VERSION"],
    azure_endpoint=environ["AZURE_OPENAI_ENDPOINT"],
    azure_ad_token_provider=token_provider,
)

@AGENT_APP.message("poem")
async def on_poem_message(context: TurnContext, _state: TurnState):
    # Configure streaming response
    context.streaming_response.set_feedback_loop(True)
    context.streaming_response.set_generated_by_ai_label(True)
    context.streaming_response.set_sensitivity_label(
        SensitivityUsageInfo(
            type="https://schema.org/Message",
            schema_type="CreativeWork",
            name="Internal",
        )
    )
    context.streaming_response.queue_informative_update("Starting a poem...\n")

    # Stream from Azure OpenAI
    streamed_response = await CLIENT.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "You are a creative assistant."},
            {"role": "user", "content": "Write a poem about Python."}
        ],
        stream=True,
    )

    try:
        async for chunk in streamed_response:
            if chunk.choices and chunk.choices[0].delta.content:
                context.streaming_response.queue_text_chunk(
                    chunk.choices[0].delta.content
                )
    finally:
        await context.streaming_response.end_stream()

OAuth / Auto Sign-In

@AGENT_APP.message("/logout")
async def logout(context: TurnContext, state: TurnState):
    await AGENT_APP.auth.sign_out(context, "GRAPH")
    await context.send_activity(MessageFactory.text("You have been logged out."))


@AGENT_APP.message("/me", auth_handlers=["GRAPH"])
async def profile_request(context: TurnContext, state: TurnState):
    user_token_response = await AGENT_APP.auth.get_token(context, "GRAPH")
    if user_token_response and user_token_response.token:
        # Use token to call Microsoft Graph
        async with aiohttp.ClientSession() as session:
            headers = {
                "Authorization": f"Bearer {user_token_response.token}",
                "Content-Type": "application/json",
            }
            async with session.get(
                "https://graph.microsoft.com/v1.0/me", headers=headers
            ) as response:
                if response.status == 200:
                    user_info = await response.json()
                    await context.send_activity(f"Hello, {user_info['displayName']}!")

Copilot Studio Client (Direct to Engine)

import asyncio
from msal import PublicClientApplication
from microsoft_agents.activity import ActivityTypes, load_configuration_from_env
from microsoft_agents.copilotstudio.client import (
    ConnectionSettings,
    CopilotClient,
)

# Token cache (local file for interactive flows)
class LocalTokenCache:
    # See samples for full implementation
    pass

def acquire_token(settings, app_client_id, tenant_id):
    pca = PublicClientApplication(
        client_id=app_client_id,
        authority=f"https://login.microsoftonline.com/{tenant_id}",
    )

    token_request = {"scopes": ["https://api.powerplatform.com/.default"]}
    accounts = pca.get_accounts()

    if accounts:
        response = pca.acquire_token_silent(token_request["scopes"], account=accounts[0])
        return response.get("access_token")
    else:
        response = pca.acquire_token_interactive(**token_request)
        return response.get("access_token")


async def main():
    settings = ConnectionSettings(
        environment_id=environ.get("COPILOTSTUDIOAGENT__ENVIRONMENTID"),
        agent_identifier=environ.get("COPILOTSTUDIOAGENT__SCHEMANAME"),
    )

    token = acquire_token(
        settings,
        app_client_id=environ.get("COPILOTSTUDIOAGENT__AGENTAPPID"),
        tenant_id=environ.get("COPILOTSTUDIOAGENT__TENANTID"),
    )

    # CopilotClient does not implement the context manager protocol.
    copilot_client = CopilotClient(settings, token)

    # Start conversation
    act = copilot_client.start_conversation(True)
    async for action in act:
        if action.text:
            print(action.text)

    # Ask question
    replies = copilot_client.ask_question("Hello!", action.conversation.id)
    async for reply in replies:
        if reply.type == ActivityTypes.message:
            print(reply.text)


asyncio.run(main())

Best Practices

  1. This SDK is async-first — use async def handlers and async with throughout.
  2. Always use context managers for clients and async credentials. Wrap every client in with Client(...) as client: (sync) or async with Client(...) as client: (async). For async DefaultAzureCredential from azure.identity.aio, also use async with credential: so tokens and transports are cleaned up.
  3. Use microsoft_agents import prefix (underscores, not dots).
  4. Use MemoryStorage only for development; use BlobStorage or CosmosDB in production.
  5. Always use load_configuration_from_env(environ) to load SDK configuration.
  6. Include jwt_authorization_middleware in aiohttp Application middlewares.
  7. Use MsalConnectionManager for MSAL-based authentication.
  8. Call end_stream() in finally blocks when using streaming responses.
  9. Use auth_handlers parameter on message decorators for OAuth-protected routes.
  10. Keep secrets in environment variables, not in source code.

Reference Links

Resource URL
Microsoft 365 Agents SDK https://learn.microsoft.com/en-us/microsoft-365/agents-sdk/
GitHub samples (Python) https://github.com/microsoft/Agents-for-python
PyPI packages https://pypi.org/search/?q=microsoft-agents
Integrate with Copilot Studio https://learn.microsoft.com/en-us/microsoft-365/agents-sdk/integrate-with-mcs

Reference Files

File Contents
references/capabilities.md Additional non-hero capabilities, operation-group coverage, and production checklists.

04 同仓库更多 Skills

microsoft
microsoft/skills

azure-advisor-cost-optimization

Analyze Azure Advisor Cost/IdleResources findings and prevalidate deletion safety, including Service Groups. WHEN: "find unused Azure resources", "what resources am I paying for but no longer using", "idle-resource savings", "historical idle-resource savings", "are these idle resources safe to delete". Generic Cost, other subcategories, and cross-category requests use azure-advisor.

★ 3.2k
microsoft
microsoft/skills

skill-creator

Guide for creating effective skills for AI coding agents working with Azure SDKs and Microsoft Foundry services. Use when creating new skills or updating existing skills.

Skill与智能体开发
★ 3.2k
microsoft
microsoft/skills

podcast-generation

Generate AI-powered podcast-style audio narratives using Azure OpenAI's GPT Realtime Mini model via WebSocket. Use when building text-to-speech features, audio narrative generation, podcast creation from content, or integrating with Azure OpenAI Realtime API for real audio output. Covers full-stack implementation from React frontend to Python FastAPI backend with WebSocket streaming.

前端开发后端与接口
★ 3.2k
microsoft
microsoft/skills

microsoft-docs

Understand Microsoft technologies by querying official documentation. Use whenever the user asks how something works, wants tutorials, needs configuration options, limits, quotas, or best practices for any Microsoft technology (Azure, .NET, M365, Windows, Power Platform, etc.)—even if they don't mention "docs." If the question is about understanding a concept rather than writing code, this is the right skill.

Skill与智能体开发
★ 3.2k
microsoft
microsoft/skills

mcp-builder

Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP), Node/TypeScript (MCP SDK), or C#/.NET (Microsoft MCP SDK).

Skill与智能体开发
★ 3.2k
microsoft
microsoft/skills

kql

KQL language expertise for writing correct, efficient Kusto Query Language queries. Covers syntax gotchas, join patterns, dynamic types, datetime pitfalls, regex patterns, serialization, memory management, result-size discipline, and advanced functions (geo, vector, graph). USE THIS SKILL whenever writing, debugging, or reviewing KQL queries — even simple ones — because the gotchas section prevents the most common errors that waste tool calls and cause expensive retry cascades. Trigger on: KQL, Kusto, ADX, Azure Data Explorer, Fabric Real-Time Intelligence, EventHouse, Log Analytics, log analysis, data exploration, time series, anomaly detection, summarize, where clause, join, extend, project, let statement, parse operator, extract function, any mention of pipe-forward query syntax.

数据分析师内容创作者写作与文案Skill与智能体开发
★ 3.2k