What is MCP and why are people in AI getting excited about it? Model Context Protocol might just be the universal connector between AI and all the apps and services we want it to use for us, that gets us to truly agentic AI. #ai #artificialintelligence #aitools #technology #tech #chatgpt #GPT #learnwithtiktok #learnontiktok #MCP #claude #agenticai

People in AI are really excited about MCP at the moment because it might just be the key to the next phase of AI, which is truly agentic AI that can just go off and do work for you without getting derailed or constantly needing babysitting, which is kind of the current state of agentic AI. So I thought I'd give you a very simple short explanation of MCP. It stands for Model Context Protocol, which doesn't really give you much clue as to what it actually is or does, but it was created by the company behind the popular LLM, Claude, to solve a really fundamental problem with agentic AI. How do we get AIs like ChatGPT and Claude to just act like agents and go off and do stuff for us, like access my email, folder storage, weather, restaurant bookings, shopping, whatever, and just do it and come back and say, job done, okay? Because we do have ways of doing that at the moment, but they are far from ideal and not really going to carry us forward. So one way that AI can currently do that is through something like ChatGPT's operator or Claude's computer use, where the AI actually goes and uses the human-designed internet, screen by screen. But that's horrendously inefficient because it's basically constantly screenshotting what's on screen and then prompting itself with those screenshots to decipher what's on screen and then work out what to do next. That's crazy inefficient. I mean, can't we just get machines talking to machines? I hear you ask. Yeah, good question. Well, that technology's existed for a very long time and that's called an API. And you use APIs all day, every day, without even knowing it. Like when you open your weather app on your phone, the weather app makes an API call to something like the weather channel and it pulls structured data, including the wind direction, the temperature, stuff like that. Now, APIs are great because they're very reliable. You set them up and they just work. But they're very inflexible. The developer of the app that you're using has to have built in the connection to the API and works out how it'll take that data and put it into the app. And that all has to be built. But that doesn't really work for AI going off around the internet and just doing stuff because those connections with the various services and sites and apps that it will use haven't been pre-programmed by a developer at some point. So that's where MCP comes in. It's like the perfect middle ground. It's the best of both worlds. It's a bit like browser use in that it's very flexible, but it doesn't have that inefficiency. And it's a bit like APIs where it can just plug into any service and it's very reliable, but without the rigidity and the need for prior programming by some developer. So you can find more out about this at Claude's website. And I guess the sort of dream of this is it becomes like HTML, MP3s, USB ports, these sort of standardized protocols that just make everything compatible with each other. Because imagine that your AI of choice, like chatGBT, you can just ask it to do a job and it will be able to go off and do absolutely anything, connect to anything, use anything, because MCP will be that universal connector. That will be when we truly live in the future depicted in the movie Her, basically.