About This Talk
Join us for a technical deep-dive meetup on the Model Context Protocol (MCP), the emerging open standard that allows LLMs to securely interact with external tools, data, and services.
We'll start by demystifying what MCP really is and how the client–server flow works, then dive straight into hands-on tooling and attack-surface experiments. You'll leave with both defensive and offensive insights into this rapidly evolving ecosystem.
What to Expect
MCP Debunked — Concise breakdown of the protocol, client ⇄ server lifecycle, transport modes and capability registration.
Debugging with MCP Inspector — Live exploration for testing, tracing and verifying MCP interactions.
Intent Analysis — Using AI to analyze tool descriptions and detect unsafe capabilities before agents can use them.
The Capability-Delta Hack Explained — Security-focused demo showing how a model can escalate or extend capabilities.