Model context protocol (MCP) servers are very buzzy right now. The idea is simple: teach large language models (LLMs) how to interact with other software systems. In doing so, LLMs can learn from and affect the real world; they can call a web service to make a phone call, invoke a command line interface (CLI) tool to add an item to a grocery list in a reminder app, and so on. To make such calls, LLMs must know what software it can call and how to do so. This is the problem that MCP solves: it in...