MCP Workshop

This project demonstrates the Model Context Protocol (MCP) with multiple server architectures and an AI-powered agent client. The implementation is available in both .NET and Python.

Architecture Overview

┌─────────────────────────────────────────────────────────────────┐
│                        MCP Agent Client                         │
│                    (Interactive AI Assistant)                   │
└──────────────┬──────────────────────────┬───────────────────────┘
               │                          │
               │ STDIO                    │ HTTP (Streamable)
               │                          │
┌──────────────▼──────────────┐ ┌─────────▼───────────────────────┐
│       MCP Local Server      │ │        MCP Bridge               │
│    (In-memory tickets)      │ │     (MCP over HTTP)             │
│         :STDIO              │ │         :5070                   │
└─────────────────────────────┘ └─────────────┬───────────────────┘
                                              │ REST API
                                              │
                                ┌─────────────▼───────────────────┐
                                │       Remote Server             │
                                │      (REST API Backend)         │
                                │          :5060                  │
                                └─────────────────────────────────┘

Components

| Component | Description | Transport | |-----------|-------------|-----------| | Remote Server | REST API backend with ticket data | Port 5060 | | MCP Bridge | MCP server that wraps REST API | Port 5070 (HTTP) | | MCP Local Server | Standalone MCP server with in-memory data | STDIO | | MCP Agent Client | AI agent that consumes MCP servers | Interactive CLI |

Available MCP Tools

All MCP servers expose the same tools for ticket management:

| Tool | Description | |------|-------------| | GetAllTickets | Gets all support tickets with optional limit | | GetTicket | Gets a support ticket by ID (e.g., TICKET-001) | | UpdateTicket | Updates a ticket's status (Open, In Progress, Resolved, Closed) |

Implementations

Choose your preferred implementation:

🔷 .NET Implementation

Built with .NET 10 and ASP.NET Core
Uses ModelContextProtocol NuGet package
Uses Microsoft.Extensions.AI for AI integration
Azure OpenAI for the agent client

🐍 Python Implementation

Built with Python 3.x and FastAPI
Uses mcp SDK for Model Context Protocol
Uses azure-ai-projects for AI integration
Azure OpenAI for the agent client

Quick Start

Each implementation includes scripts to launch all services:

Windows

# For .NET
cd dotnet
start_all.cmd

# For Python
cd python
start_all.cmd

macOS / Linux

# For .NET
cd dotnet
chmod +x start_all.sh   # Make executable (first time only)
./start_all.sh

# For Python
cd python
chmod +x start_all.sh   # Make executable (first time only)
./start_all.sh

See the implementation-specific README for detailed setup instructions.

Shared Data

All implementations read ticket data from a shared JSON file:

data/tickets.json

This ensures consistent data across .NET and Python implementations.

Exercises (Optional)

Ready to explore more? Check out EXERCISES.md for hands-on tasks to extend the workshop:

Add new MCP tools (CreateTicket, DeleteTicket)
Implement ticket filtering and search
Add authentication to the MCP Bridge
Extend the data model with comments
And more bonus challenges!

License

MIT

MCP Servers