《Can’t tell the difference between RAG, MCP, Agent and AI concepts? This article will turn newbies into experts in seconds! 》

Let’s explore the three best friends of the AI ​​world: MCP, RAG, and Agent! Today we will take you through these exciting technologies in an easy way.

These three friends have become very popular in the AI ​​community recently! Especially MCP, which brings a lot of interesting new ways to play, making everyone eager to try it out~

However, I know that some of you may find these concepts a little difficult to understand. Don't worry, this is normal! After all, the current materials are written in a more "serious" way and may not seem very user-friendly.

So today, we plan to talk to you in a simple and easy-to-understand way about how these three AI friends work together and what interesting things they can each help us do.

Are you ready? Let's learn about these powerful technologies in a relaxed and enjoyable way!

Let’s talk MCP!

Friends, there are two different "families" of MCP that we are going to learn about today:

One is Model Compression and Pruning, and the other is Model Context Protocol. These two sound very professional, right? Don't worry, let's take a look~

First, let’s talk about model compression and pruning. It is like “losing weight” for AI models! It helps machine learning models become slimmer and more efficient while maintaining their original strength.

But let’s not discuss this topic in depth today, we’ll leave it for next time!

The Model Context Protocol is a great open protocol launched by Anthropic, the parent company of Claude, in November 2024. Imagine that it is like installing a magical translator on the AI ​​model, allowing AI to communicate freely with various tools and data sources!

Simply put, it is a treasure chest of AI big models!

With this treasure box, AI can happily interact with the outside world, obtain information, and complete various interesting tasks.

Remember when we wanted to share files with AI, we always had to copy and paste manually? It was so troublesome, right?

Now with MCP, a thoughtful little assistant, AI can directly access and use various tools, such as browsers, files, etc., which is much more convenient!

(Let me give you an interesting analogy: MCP is like a super universal converter. Whether it is a computer file, a company database, or WeChat/DingTalk/GitHub, it can be connected with one click, saving a lot of trouble!)

Some of you may ask: Hey, isn’t this similar to function call? Let me explain.

Function call is like a little helper for AI, helping it complete some specific tasks.

For example, if you ask "What's the temperature in Beijing today?", it will check the weather and tell you the answer.

However, although function call is very powerful, it is a bit of a loner and can only play in its own yard.

MCP is really powerful. It unifies the "dialects" of all large AI models into a kind of "Mandarin", allowing everyone to communicate happily!

After reading this, I believe you have a better understanding of MCP!

rag

Speaking of RAG, it's like a "super library" for our AI assistant!

Imagine that when we ask AI questions, it doesn't make assumptions out of thin air, but goes to this "library" to look for information first, just like a serious primary school student doing homework. This way the answer will be more accurate and not made up!

The way it works is particularly interesting: when you ask a question, it will be like a diligent little librarian, first turning your question into a special "search symbol", and then running to the bookshelf (that is, the knowledge base) to find relevant content. After finding it, it will organize the information and use it to answer your question, just like a careful teacher!

What interesting things can RAG do for us? Let's take a look~

Smart Assistant

• The customer service elf can quickly browse the product manual and give you the most considerate answer

• The warm-hearted medical assistant will carefully review the medical records and give professional advice

Enterprise Data Manager

• Help colleagues easily find various company documents and guidelines, saving the trouble of searching everywhere

Smart Decision Assistant

• Help financial analysts organize market data and write beautiful investment reports

• Assist legal advisors to find similar cases and analyze possible risks

Agent - Our AI assistant!

Friends, this is a super interesting topic that we have been talking about a lot recently! If you want to learn more about this lovely AI assistant, please visit my homepage to check out the Intelligent Agent column~

Now let’s take a look at how these three good friends play together!

MCP and RAG's sweet partnership (a perfect combination of knowledge and execution)

Imagine that MCP is like RAG's right-hand man, helping it find the tools it needs!

RAG is like a little bookworm who loves to learn and needs various tools to find knowledge, while MCP is like a caring housekeeper who puts all the tools neatly in place and is always ready to help.

For example, when Agent wants to search for information online, it will ask MCP to help connect to the search tool, and then hand over the found content to RAG to organize and answer, just like a warm little team!

And RAG will in turn help MCP become smarter~

When MCP is handling various tasks, RAG is like a treasure trove of knowledge, providing it with the information it needs at any time.

Let’s take a small example: when managing product inventory, MCP can easily view real-time inventory data, just like having a portable assistant!

Let’s use a warm and interesting example to see how these three good friends work together through Coze’s AI popular article assistant.

First is the little helper RAG: when you ask a question, it is like a hardworking little bee, flying to the knowledge base garden to collect information (such as product manuals).

Then there is MCP, a little helper: if you need to do something specific (such as placing an order), it is like a flexible elf that helps you open various tools, generate charts or process data.

What is the relationship between MCP and Agent? They are like a pair of super tacit partners!

Just like dancing a duet, they work together very well: one is responsible for standardizing the movements (MCP), and the other is responsible for performing wonderful dance steps (Agent).

Through such tacit cooperation, they make AI smarter and more capable!

MCP is like an Agent's treasure chest

It neatly stores all kinds of tools, waiting for the Agent to use them at any time.

For example, if the Agent wants to check the database or send an email, he can easily find the required tools from the treasure box.

Agent is like a clever little ghost

It needs the help of MCP to complete the task. For example, when it wants to check the weather, it will ask MCP to help open the weather forecast tool.

Sounds a bit abstract? No problem, let's use a small example in real life to illustrate.

Suppose you have a super smart washing machine at home, and you just say: "I want to wash clothes!"

At this time, the smart washing machine (our Agent) will act like a housekeeper and divide the task into several small steps:

→ First use your eyes to see what the clothes are made of (ask the "Fabric Identification Assistant" for help)

→ Think again about how hot the water should be (ask the "water temperature adjustment assistant" for help)

→ Finally, start washing clothes (ask the "Laundry Program Assistant" for help)

The MCP is like a magical master switch, responsible for connecting all these assistants:

Fabric recognition assistant connected to "smart camera"

The water temperature adjustment assistant is connected to the "heartwarming water heater"

The laundry program assistant is connected to the "hard-working little motor"

MCP is like an enthusiastic electrician who helps to connect all the tools but won’t do the laundry himself.

The Agent is like a caring butler who knows when to use which tool and, with the help of MCP, washes clothes cleanly!