Turn Claude into a Real Assistant with MCP

This introductory video from Relay explains the fundamentals of tool calling and the Model Context Protocol (MCP) through a practical use case: building an AI assistant that cancels meetings and drafts apology emails when a child is sick. The first half provides clear conceptual explanations of how large language models work, why tool calling is necessary for LLMs to take real-world actions, and what MCP actually is—not a tool itself, but a standardized protocol that makes it easier for tools to communicate their capabilities to LLMs, analogous to standardized restaurant menus.

The second half is a step-by-step walkthrough of creating an MCP server in relay.app with two tools (calendar event lookup and email draft creation), connecting it to Claude as an MCP client, and demonstrating the complete workflow. The video shows real-time debugging when the initial implementation fetches too many calendar events, illustrating the iterative refinement process. For product managers new to MCP and tool calling, this provides the foundational understanding needed to evaluate AI assistant architectures and participate in technical discussions about agentic system design.