MCP Workshop

Table of Contents

1. Introduction to MCP
2. Core MCP Concepts
3. Workshop Setup
4. Console Client
5. EntraID Client
6. Demo 1: Basic Stock Server
7. Demo 2: OAuth-Protected Weather Server
8. Demo 3: OAuth 2.0 On-Behalf-Of MCP Server
9. Demo 4: Secure Access to MCP Servers in Azure APIM
10. Demo 5: MCP with Azure AI Foundry
11. Demo 6: MCP Bindings for Azure Functions
12. Demo 7: Logic Apps Connectors as MCP Servers using Azure API Center
13. Demo 8: Extend Copilot Studio Agent Using MCP
14. Demo 9: Private and Enterprise-Ready MCP Registry
15. Security Risks in MCP
16. Demo 10: XPIA Attacks in MCP
17. Demo 11: Remote Code Execution (RCE) in MCP
18. Demo 12: MCP Tool Poisoning
19. Demo 13: MCP Tool Shadowing
20. Hands-on Exercises
21. Key Takeaways
22. Resources
23. Next Steps

Introduction to MCP

The Model Context Protocol (MCP) is an open standard that enables AI assistants to securely connect to data sources and tools. Think of it as a universal API that allows language models to interact with external systems in a standardized way.

Why MCP?

• Standardization: One protocol for all integrations
• Security: Built-in authentication and authorization
• Flexibility: Supports multiple transport methods
• Extensibility: Easy to add new capabilities

Core MCP Concepts

1. Resources

Resources are read-only data sources that can be accessed by AI models. They represent information that the model can read but not modify.

Example from our stock server:

@mcp.resource("stock://{symbol}")
def stock_resource(symbol: str) -> str:
    """Expose stock price data as a resource."""
    price = get_stock_price(symbol)
    return f"The current price of '{symbol}' is ${price:.2f}."

2. Tools