Azure AI Foundry Agent / MCP Server / MS Teams Bot Project (Beta/Preview SDKs)

A learning project demonstrating integration between Azure AI Foundry models/agents, MCP (Model Context Protocol) servers, and Microsoft Teams bots using preview and beta versions of various SDKs.

Please see my LinkedIn posts for more info:
https://www.linkedin.com/feed/update/urn:li:activity:7365798255726579714/
https://www.linkedin.com/feed/update/urn:li:activity:7378160986584670208/

Overview

This project showcases seven different AI interaction approaches:

• Base Model - Azure AI Foundry model with full conversation history
• Goldfish Model - Azure AI Foundry model with no memory
• Claude Code - Local Claude Code with local Enphase MCP Server (stdio transport)
• Azure Agent - Azure AI Foundry Agent without MCP tools
• MCP Agent - Azure AI Foundry Agent with Enphase MCP Server access (HTTP transport via ngrok)
• Local Foundry - Local Azure AI Foundry model with full conversation history
• Local Foundry Goldfish - Local Azure AI Foundry model with no memory

The Teams Bot serves as a unified interface to interact with all these AI variants through simple command prefixes.

Project Structure

Quote.Agent (Main Teams Bot)

• Program.cs - ASP.NET Core startup configuration with Teams middleware
• MainController.cs - Teams bot controller handling message routing and AI integrations
• Supports all seven AI interaction modes with proper resource cleanup using the Dispose pattern

TestMcpAgent (Standalone Console App)

• Program.cs - Demonstrates Azure AI Foundry Agent with MCP Server integration
• Two-part conversation example for testing agent functionality
• Comprehensive error handling and resource cleanup

TestLocalFoundryModel (Standalone Console App)

• Program.cs - Demonstrates Azure AI Foundry Local model integration
• Shows local model startup, inference, and management
• Uses phi-4-mini model for hardware compatibility

Key Features Demonstrated

Azure AI Foundry Integration

• Base model chat with conversation history
• Local model integration using Azure AI Foundry Local
• Agent creation and management
• Thread and message handling
• Asynchronous operations with proper polling

MCP Server Integration

• HTTP transport protocol with authorization
• Stateless server configuration
• Tool definition and approval workflows
• Dynamic header management for authentication
• Optional tool filtering (allow specific tools only)

Teams Bot Functionality

• Message routing based on command prefixes
• Multiple AI backend support
• Typing indicators and proper message formatting
• Integration with local Claude Code via WSL

Authentication & Authorization

• Azure CLI credential integration
• JWT token management for MCP server access
• Authorization header handling for tool approvals
• Support for both automatic and manual tool approval flows

Prerequisites

• .NET 9.0 (Quote.Agent, TestMcpAgent)
• .NET 10.0 (TestLocalFoundryModel)
• Azure CLI (az login required)
• Azure AI Foundry project with deployed model
• Azure AI Foundry Local (for local model support)
• Enphase MCP Server running locally
• ngrok for exposing local MCP server to Azure (for HTTP transport)
• Microsoft Teams for bot testing

Configuration

The project uses appsettings.json and appsettings.development.json for configuration:

{
  "ProjectEndpointForModel": "your-azure-ai-foundry-model-endpoint",
  "ProjectEndpointForAgent": "your-azure-ai-foundry-agent-endpoint", 
  "ModelDeploymentName": "your-model-deployment-name",
  "ApiKey": "your-api-key",
  "McpServerUrl": "your-ngrok-url-for-mcp-server"
}

Usage

Teams Bot Commands

Start your message with one of these prefixes:

• model - Chat with base model (full conversation history)
• goldfish - Chat with model (no memory)
• `

... View full README on GitHub