MCP: What is Model Context Protocol

The open standard that lets your AI tools talk to everything else.

Before MCP, every AI coding agent had its own bespoke way of plugging into databases, APIs, ticketing systems, and whatever else you wanted it to touch. Anthropic released Model Context Protocol as an open standard in late 2024 to fix that: one protocol, thousands of servers, any compliant client can use any compliant server. If you’ve used Claude Code with the GitHub, Postgres, or filesystem servers, you’ve already been using MCP — this session is about understanding what’s actually going on and using it deliberately. Aimed at anyone who uses Claude Code or a similar agent for real work and wants to extend what it can do beyond the built-in tools. If you’ve ever wished your coding agent could query your production database, file your Linear tickets, or read your Notion docs without you pasting in results, you’re the target audience. By the end of the 90 minutes you will: • Know what MCP actually is, how servers and clients talk to each other, and why "one protocol to rule them all" matters. • Have added two or three MCP servers to your own agent — at minimum filesystem and GitHub — and used them live. • Know where to find the servers that already exist and how to evaluate whether one is safe to use. • Understand the real security risks: credential exposure, prompt injection via server responses, over-privileged tool grants — and how to mitigate each. How the session runs: 20 minutes on what MCP is and how it works under the hood — JSON-RPC, tools vs resources vs prompts, stdio vs HTTP transports. Then we add servers to your Claude Code setup together, run real queries, and see the agent use them. Back half: we go through the security model — what a malicious or buggy MCP server can actually do to you, and what to check before installing one. Bring your laptop, charged. Have Claude Code installed and working.