Model Context Protocol (MCP) is quickly becoming the standard way AI agents connect to tools, data, and external systems. But after the hype fizzles, you may be left with unanswered questions: When models improve every week, why introduce a new protocol at all? Aren’t CLIs or API wrappers enough?

Most explanations focus on what MCP does without explaining why it exists. In this talk, MCP contributor, OAuth educator, and Arcade.dev founding engineer Nate Barbettini builds a clear mental model of MCP from first principles. Drawing on years of explaining protocol design and security systems to engineers, he’ll start with the real constraints agents face, then build upward to protocol capabilities and the new patterns of agent interaction.

Along the way, he'll untangle the OAuth acronym soup (PKCE, DCR, CIMD, oh my!) and demystify how secure agent connections actually work.

If you think MCP is just a thin API wrapper, this talk will change how you see it. You’ll leave with clarity: what MCP is, why it exists, and how to use it correctly.
Sobre Nate Barbettini: Nate Barbettini is a founding engineer at Arcade.dev, building the MCP runtime for secure, multi-user AI agents. Previously, he was co-founder and CTO of Cobbler, where he built production AI systems for finance teams. A longtime educator in authentication and security, Nate actively contributes to MCP and AI security projects. He lives in the Bay Area with his wife and two curious kids. Show less