All things MCP

The Model Context Protocol (MCP) has quickly become the standard way to connect AI models to your tools, data, and services — and if you're building AI applications in 2026, it's a skill you can't skip.
In this session we'll take a practical, code-first tour of the MCP ecosystem:

Connecting to existing MCP servers — We'll hook an AI application up to off-the-shelf MCP servers (filesystem, GitHub, databases, and more) and show how tools, resources, and prompts are discovered and invoked at runtime.

Building your own MCP server — We'll write one from scratch, covering both transport options: stdio for local integrations, and streamable HTTP for remote, multi-client deployments. You'll see how to define tools, handle requests, and test your server before wiring it into a client.

Security — MCP gives models real capabilities, which means real risk. We'll cover authentication and OAuth for remote servers, prompt injection and tool poisoning attacks, permission scoping, and how to avoid the "confused deputy" problem when your server acts on behalf of users.

Plus: the current state of the MCP spec, registries and discovery, debugging with the MCP Inspector, and where the protocol is headed next.
Whether you're integrating existing servers or shipping your own, you'll leave with working code and a clear mental model of how MCP fits into modern AI application architecture.

Beginners welcome — no prior MCP experience required. Bring a laptop if you want to follow along.