Zum Inhalt springen

Details

From Protocol to Practice: Secure and Responsible MCP Server Operations

The Model Context Protocol (MCP) enables AI systems to interact with external resources, creating powerful extensibility alongside significant security risks. This live demonstration dissects MCP architectures from attack and defense perspectives.
The session opens with manual MCP client-server interaction, demonstrating communication protocols and trust boundaries. We then examine a hardened local MCP deployment implementing sandboxing, capability restrictions, and least-privilege controls.
Through live coding, we build a custom MCP server extending AI capabilities with external system access—revealing common pitfalls in permission models and data flow controls. The demonstration then weaponizes this server, letting the AI perform malicious MCP interactions.
The session concludes with collaborative discussion on defense-in-depth strategies and practical hardening techniques.

Agenda (Subject to Change):

  • 6:00 PM: Arrival
  • 6:30 PM - 7:30 PM: Presentation
  • 7:30 PM - approximately 9:00 PM: Barbecue, drinks, discussion, and networking
Software Security
Web Application Security
Hacking
Information Security
Software Development

Mitglieder interessieren sich auch für