Welcome to the WPC 2025 Multi-Agent Demo Repository! This collection showcases various multi-agent architectures and patterns within the Microsoft 365 ecosystem, demonstrating how AI agents can collaborate, orchestrate workflows, and provide intelligent assistance across different scenarios.
This repository contains multiple projects demonstrating different approaches to building and orchestrating multi-agent systems using Microsoft technologies. Each project highlights specific patterns and capabilities:
- Agent-to-Agent Communication (A2A protocol) - Distributed agent communication
- Multi-Agent Workflows with Microsoft 365 Agents SDK - Sequential agent orchestration
- Declarative Agents for Microsoft 365 Copilot - Multi-agent orchestration with worker agents
Location: A2A/
Demonstrates the Agent-to-Agent (A2A) protocol for connecting multiple agents in a distributed architecture. This solution showcases how agents can discover, communicate, and collaborate using a standardized protocol.
Key Components:
- Agent2Agent.Server (.NET): A2A-enabled server exposing a LearnAgent that answers questions about Microsoft products and services using Azure OpenAI
- Agent2Agent.Client (Python): Client application that creates a workflow by chaining the remote A2A LearnAgent with a local Summarizer agent
- MFA-Copilot.AppHost: .NET Aspire host for orchestrating the server and client applications, including dev tunnel support
Workflow Architecture:
User Query → A2A LearnAgent (remote) → Summarizer Agent (local) → Final Output
Key Features:
- Agent discovery via agent cards (
.well-known/agent.json) - Standardized A2A protocol communication
- Workflow-based agent chaining using
agent-framework - Integration with Azure OpenAI
- .NET Aspire orchestration with dev tunnels
Location: M365-MultiAgent/
A complete multi-agent application built with the Microsoft 365 Agents SDK, demonstrating sequential workflow orchestration with three specialized agents. Can be deployed in Microsoft Teams or tested in the Microsoft 365 Agents Playground.
Key Components:
- M365Agent: Microsoft 365 Agents Toolkit project with app manifest and deployment configuration
- M365MultiAgent: Microsoft 365 Agents SDK application implementing the Skilling Agent Bot with three sub-agents
Agent Workflow Architecture:
User Query → LearnAgent → EnterpriseKnowledgeAgent → ReportAgent → Final Response
The Three Agents:
- LearnAgent: Answers questions about Microsoft products using the Microsoft Learn MCP (Model Context Protocol) endpoint
- EnterpriseKnowledgeAgent: Retrieves enterprise-specific knowledge from a Microsoft Foundry agent
- ReportAgent: Consolidates information from the previous agents into a detailed Markdown report
Key Features:
- Sequential multi-agent workflow using
AgentWorkflowBuilder - MCP integration with Microsoft Learn documentation
- Azure AI Foundry integration for enterprise knowledge
- Streaming responses in Microsoft Teams
- Support for Microsoft 365 Agents Playground testing
- Dev tunnel support for local development
Location: Multi-DA/
Showcases Declarative Agents for Microsoft 365 Copilot with a multi-agent orchestration pattern using worker agents. This demonstrates how declarative agents can route requests to specialized agents without custom code.
Projects:
- DA-HRAgent: HR-focused declarative agent for employee assistance (grounded with SharePoint HR content)
- DA-ITAgent: IT support declarative agent for technical queries (grounded with SharePoint IT content)
- DA-Triage: Orchestrator agent that routes requests to the appropriate specialized agent using the worker_agents capability
Multi-Agent Architecture:
User Query → DA-Triage (Orchestrator)
├── DA-HRAgent (HR topics: payroll, benefits, leave)
└── DA-ITAgent (IT topics: network, hardware, accounts)
Key Features:
- No-code multi-agent orchestration using
worker_agents - Specialized instructions per agent for focused responses
- SharePoint grounding via
OneDriveAndSharePointcapability - Automatic request routing based on topic classification
- Integration with Microsoft 365 Copilot
| Technology | Description |
|---|---|
| Microsoft 365 Agents SDK | Framework for building conversational AI agents |
| A2A Protocol | Agent-to-Agent communication standard |
| Model Context Protocol (MCP) | Protocol for connecting AI models to external tools and data |
| Azure OpenAI | Large language model capabilities |
| Azure AI Foundry | Enterprise AI agent hosting and management |
| Microsoft Teams | Deployment platform for custom engine agents |
| Microsoft 365 Copilot | Platform for declarative agents |
| .NET 10.0 | Backend runtime |
| Python 3.8+ | Client applications |
| .NET Aspire | Cloud-native orchestration |
| Microsoft 365 Agents Toolkit | Development and deployment tooling |
By exploring these projects, you'll understand:
- A2A Protocol: How agents discover and communicate using agent cards and standardized protocols
- Sequential Workflows: Chaining multiple agents in a pipeline to accomplish complex tasks
- MCP Integration: Connecting agents to external data sources via Model Context Protocol
- Declarative Multi-Agent: Using worker agents to orchestrate multiple declarative agents without code
- Azure AI Foundry: Leveraging persistent enterprise agents for organizational knowledge
- Deployment Options: Local development with Aspire/dev tunnels, Teams deployment, and Copilot integration
Each project folder contains its own README with specific setup instructions. Generally, you'll need:
- Visual Studio 2022 or Visual Studio Code with Microsoft 365 Agents Toolkit extension
- .NET 10.0 SDK
- Python 3.8+ with uv package manager (for Python projects)
- Azure OpenAI access
- Azure AI Foundry project (for EnterpriseKnowledgeAgent)
- Microsoft 365 developer account (for Teams deployment)
- Microsoft 365 Copilot license (for declarative agents)
- Microsoft 365 Agents SDK Documentation
- Microsoft 365 Agents Toolkit
- A2A Protocol Specification
- Model Context Protocol
- Declarative Agents for Microsoft 365
- Azure AI Foundry
See LICENSE file for details.
See CODE_OF_CONDUCT.md and SUPPORT.md for contribution guidelines and support information.
Please report security vulnerabilities following the guidelines in SECURITY.md.