atul |
model context what?
Yup, model context protocol; sounds similar to MVP (model, view, presenter), MVC (model, view, controller), doesn’t it? Well, that’s exactly what it is, a framework, bunch of specifications on how to code a certain thing. So what is it for? well it is mostly used for AI apps to build a generic adapter layer between the client and server which can server a single interface for integrating mutiple tools.
Basically using MCP you can give your AI agent context about certain tool and then you could implement the certain tool based on some specs and priciples; this is the overall idea. Didn’t langchain built the same thing? well, yes and no; its same but different; lol.
There are all bunch of abstractions for solving same problem that is adapters for the AI agent to use. It’s just sounds fancy since it uses the three letter short-form.
why so popular?
The entire AI twitter is hyping this MCP thing, it’s not even new; it was made public by some company in 2014; people are just jumping on the hype train again and calling it the next big thing.
I may be the biggest hater of using OpenAI APIs in the backend and calling it AI app; using this new framework won’t make it more fancier. I will say, it would be less innovative since you are tying yourself down with the framework.
should i use it?
Depends, will it solve anything for you? If you are building an app using LLMs you probably have something similar setup. Your current project is probably doing the same thing; giving LLM the context it needs to produce reasonable output. So, if it does not fit into yoru workflow no point reading the specs.
If you haven’t started yet, probably give it a shot and then have hateful things to say about it.
better responses
If you are thinking of getting better reponses from LLM just by using this unnecessary-bloated abstraction layer then you are mistaken, this is not going to solve the hallucination problem, and that too for a LLM which is trained for half of the worlds’ dummest living being.
summary
Do your own thing, research and learn. Instead of using the “next big thing”, try to build something yourself, may be look at the source code and see whether it solves anything real for you.
Well, most of the AI stuff out there is mediocre; bunch of API calls and unnecessary data hoarders. It’s just about jumping the hype train while it’s running and making bank!
Do you know, people who solve unnecessary problems are called “architecture astronauts”