m365-agents-py_skill

Overview

This skill provides a Python SDK wrapper and examples for building Microsoft 365, Teams, and Copilot Studio agents using aiohttp hosting, AgentApplication routing, streaming responses, and MSAL-based authentication. It surfaces the microsoft_agents import prefix, connection patterns, and ready-to-use handlers for messages, invokes, and conversation updates. The package emphasizes secure environment configuration and production-ready storage/auth patterns.

How this skill works

The SDK wires an AgentApplication to a CloudAdapter and an aiohttp server endpoint, routing incoming Activities to decorated handlers (message, activity, conversation_update). It manages authentication via MsalConnectionManager, supports auth-protected message handlers with auth_handlers, and enables streaming responses by integrating Azure OpenAI async clients into context.streaming_response. Utility functions load configuration from environment variables and provide middleware (jwt_authorization_middleware) for request validation.

When to use it

• Build conversational agents for Microsoft Teams and Microsoft 365 using Python.
• Host agents that require MSAL-based OAuth flows and Graph API calls.
• Stream LLM-generated content (Azure OpenAI) into Teams or Copilot Studio sessions.
• Route activities with regex or string-based message handlers and add auth-protected endpoints.
• Integrate direct-to-engine Copilot Studio flows with a CopilotClient.

Best practices

• Use the microsoft_agents import prefix (underscores) to match current package structure.
• Keep secrets in environment variables and load config with load_configuration_from_env(environ).
• Use MemoryStorage only for development; switch to Blob/Cosmos storage in production.
• Include jwt_authorization_middleware in aiohttp middlewares for secure endpoints.
• Always call context.streaming_response.end_stream() in finally blocks when streaming.
• Protect OAuth routes with auth_handlers and use MsalConnectionManager for token management.

Example use cases

• Echo bot: lightweight message handler that echoes user input via AgentApplication.activity("message").
• Welcome flow: conversation_update handler to greet new members when added to a chat or team.
• Profile lookup: /me command protected by GRAPH auth_handlers to call Microsoft Graph and show the user profile.
• Streaming assistant: stream a long-form response from Azure OpenAI into Teams with informative updates and sensitivity labels.
• Copilot Studio integration: use CopilotClient to start conversations and relay engine replies to a user session.