LLM, RAG, AI Agents and MCP. What's the Difference?
AI has brought quite a few new terms into our everyday conversations. LLM, RAG, AI Agents, MCP and plenty more. We keep coming across these terms in articles, product announcements and discussions about what AI can do.
লেখক Anjan N Ghosh··49K বার পড়া
LLM, RAG, AI Agents and MCP. What's the Difference?
AI has brought quite a few new terms into our everyday conversations. LLM, RAG, AI Agents, MCP and plenty more. We keep coming across these terms in articles, product announcements and discussions about what AI can do.
But what do they actually mean? And how are they different from one another?
Here’s a simple way to understand them.
LLM (Large Language Model)
This is what powers conversational AI tools such as ChatGPT. You ask a question, and the model generates a response using what it learnt during training and the information available in your conversation.
It can draft an email, summarise a document, explain a subject or help you work through a problem. But it doesn’t automatically know everything happening around us, and sometimes it can give a completely wrong answer with remarkable confidence.
Think of an LLM as someone who has read an enormous number of books and can discuss almost any subject. That doesn’t necessarily mean everything they say is correct.
RAG (Retrieval-Augmented Generation)
Now imagine asking that same person a question about your company’s leave policy.
Rather than relying on memory, they open the latest HR document, find the relevant section and answer your question.
That’s roughly how RAG works.
It allows an AI system to search documents, databases and other information sources before generating an answer. This is particularly useful when the information is specific to an organisation or changes regularly.
For example, an employee could ask about a company policy, and the AI could retrieve the relevant information from the organisation’s documents instead of relying entirely on its training.
Of course, if the source document is outdated or incorrect, the answer could still be wrong.
AI Agents
This is where AI goes beyond answering questions.
Suppose you’re planning a holiday. You ask ChatGPT to suggest a destination, compare hotels and prepare an itinerary. That’s something a regular AI assistant can help with.
An AI Agent can go further. With access to the necessary tools and your permission, it could search for flights, compare prices, check availability and carry out parts of the booking process.
In other words, instead of simply telling you how to do something, it can take steps towards doing it.
That’s also why permissions matter. You wouldn’t want an AI Agent spending money, sending emails or making bookings without your approval.
MCP (Model Context Protocol)
MCP is slightly different from the other three.
It’s a standard that helps AI applications connect with external tools and information sources.
For example, suppose you want an AI assistant to check your calendar, read a project document and prepare a meeting summary.
Those details may be stored in different applications. MCP provides a common way for compatible AI systems to connect with those services.
It doesn’t generate answers or make decisions by itself. It simply helps establish the connections through which AI applications can access tools and data.
How do they work together?
Let’s take a simple office example.
You ask an AI assistant to prepare a report on the progress of an ongoing project.
The LLM understands your request and helps write the report. RAG retrieves relevant information from project documents. An AI Agent can organise the steps and use available tools to prepare the report. MCP may provide the connections to the applications where the information is stored.
Not every AI system needs all four, and they don’t always work in this exact sequence.
But the basic difference is easy enough to remember.
LLM generates responses. RAG retrieves relevant information. AI Agents carry out tasks. MCP connects AI applications with tools and data.
The next time you come across these terms in an article or product announcement, you’ll have a better idea of what each one actually does.