What exactly is an MCP server?
A simple explanation of what it is and what it's used for
Welcome to Infinite Curiosity, a newsletter that explores the intersection of Artificial Intelligence and Startups. Tech enthusiasts across 200 countries have been reading what I write. Subscribe to this newsletter for free to directly receive it in your inbox:
"Can you explain MCP server to me? And speak to me as you might to a young child or a golden retriever."
You've certainly heard of MCP server by now. And someone recently told me they don't know what it is and now they're too afraid to ask. Obviously they were being facetious, but it got me thinking. This is my attempt to explain it in a way I want to be explained.
Let's say you have a tool or a model that does something specific. An MCP server is a web server to turn this tool / model into an "LLM-accessible function".
It exposes a structured json API describing and implementing tools/functions that an LLM can call over the internet. If you want humans to ping your tool, you'll wrap it in a REST API. If you want LLMs to ping your tool, you'll need to wrap it in an MCP server.
Here's an example:
You have a weather API. And you want ChatGPT to call it when a user asks "What's the weather in San Francisco?"
You build an MCP server like:
- Have a json file that declares a weather function
- An API endpoint that accepts json with a city name like {"city": "San Francisco"}
- Returns a json like {"temperature": 72, "unit": "Fahrenheit", "condition": "Sunny"}
ChatGPT can now call your API like a function using structured JSON. You don't need to publish a plugin or build a UI.
If you're a founder or an investor who has been thinking about this, I'd love to hear from you.
If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 friend who’s curious about AI: