say less, fam: an mcp explainer

Everyone has likely heard the term MCP and knows at a surface level that it’s the standard for tool communication with AI Agents - but there’s a difference between knowledge & understanding. For me, I knew this to be the case, but I didn’t really comprehend it until I was able to wrap it in a silly little analogy.

However, quick disclaimer: this is not a technically perfect explanation, this is just the concept that helped me make sense of it and understand the implications. I have no doubt that there are some technical inconsistencies - but the point isn’t to technically explain, it’s conceptual. Anyway, hope this helps!

Think of your AI agent as a phone.

In the EU, every phone must have a USB-C port now, it’s the standard way of connecting things to your phone. It doesn’t matter what you’re plugging in, the port is always the same shape and always speaks the same language.

That standard is roughly what MCP is; it’s the agreed way that your agent connects to outside tools.

Now, the tools themselves; essentially things that you plug into your phone; a charger, headphones, maybe a card reader, or a little scanner, whatever - could be anything. Each one does a specific thing, but because your phone uses a standard port/connection it doesn’t need to know how any of them work on the inside - all it needs to be able to do is talk to the device.

So you plug something into your phone and your phone goes “who are you and what can you do?” and a card reader would say something like “I read cards, and process payments, throw me some plastic and I’ll show you” (the card reader is not very professional). Now the phone knows that it can ask the card reader to process payments.

In practice, if you’re charging someone for a purchase from your phone, you enter the value, press checkout, then the phone asks the card reader “process that payment please” - the reader goes “say less fam,” reads the card, processes the payment, and sends the result. The phone has no idea how to read a card itself, and it doesn’t need to. It just knows how to ask, and the tool knows how to do the thing and hand back the answer.

Swap the card reader for a scanner, a printer, a barcode scanner, and nothing about how the phone asks changes. Same port, same language, different tool. You can keep plugging in new capabilities without teaching the phone anything new.

That, roughly, is MCP. One standard way to plug stuff into an AI so it can have more capabilities. I’m sure that there’s probably a better way of explaining this - but I’m a simple, simple man who sometimes needs to put things into words a 5-year-old can understand for it to all make sense.