Hook your LLM to the real world: building an MCP server from scratch

LLMs can read web pages, but reading isn't doing! Closing that gap between knowing and acting, and reaching private data too, is exactly why the Model Context Protocol (MCP) exists. In this talk, we'll build an MCP server from scratch and expose real tools, resources and prompts that any agent can discover and call on its own.

Along the way, we'll look at what separates a server that works in a demo from one that holds up in the real world: exposing capabilities an agent can actually use, keeping things secure and reliable, and making it portable rather than tied to a single tool. You'll leave knowing when MCP genuinely beats a plain API call or a URL, the traps to dodge, and how to plug your own tools and data into any agent.