A beginner's guide to MCP for ordinary people

I have been so busy lately that as a programmer, I haven't really learned much about MCP. Today, I finally had some time to "chew" MCP. Here I will explain to you what MCP is in plain language~

What is MCP?

1.1 Basic Concepts

The core goal of MCP (Model Context Protocol) is to solve the communication problem between large language models (LLM) and external tools.

1.2 Popular explanation

An example?

Your smart speaker will control home appliances such as air conditioners

• The air conditioner said: "Use infrared to send commands"
• The camera said: "Use Wifi to send commands"
• The robot vacuum cleaner said: "Use Bluetooth to send and receive commands"

At this time, MCP stood up and said, "Stop arguing! Speak your own dialects and I'll be the translator!"?

Without MCP, the speakers would have to learn the protocol of each device in order to communicate with them.

With MCP, the speaker only needs to "talk" to MCP, which will be responsible for translation, while the interaction between MCP and the device will be implemented by each device manufacturer.

1.3 The role of MCP

1. Unified Interface

• Just like USB-C chargers can charge various devices
• AI only needs to learn one "language" to call all tools

• Plug and Play
• New tools can be used directly by connecting to MCP
• No need to modify the AI ​​system itself
• Active execution
• AI big models can call tools to execute users' complex needs
• Safe and reliable
• Built-in permission control
• Support enterprise-level data protection
• Real-time interaction
• Support real-time data streaming
• Such as live broadcast data, sensor data, etc.

1.4 A concrete example

Imagine you are using AI customer service:

1. You ask "Where is my package?"
2. AI connects to the logistics system through MCP query
3. The map API is called through MCP to display the location
4. If you want to change the delivery address, it can also connect to the order system through MCP to modify all these operations. There is no need for AI to "learn" the specific language of each system in advance. MCP relies entirely on silent translation in the background.

This is the magic of MCP: it makes AI truly a "universal assistant" that can use various tools as flexibly as humans to complete tasks!

MCP system architecture

MCP adopts a "client-server" architecture, which includes three core roles:

• MCP Host
• MCP Client
• MCP Server

Imagine MCP is like an efficient takeout ordering system:

2.1 MCP Host - You (Customer)

Who is it: It is you, or the tool you are using to order food, such as the food ordering app on your phone (this app is equivalent to the platform running AI, such as the Claude chat window or a programming software).

What it does: You have a need (you are hungry and want to eat), you order food (issue instructions) through the App (host). It initiates a request, saying "I want a serving of Kung Pao Chicken".

What’s inside: The app integrates a small assistant (MCP client) that is responsible for contacting merchants and running errands.

2.2 MCP Client - "Contact the Running Assistant" in the App

Who is it: It is hidden in your food delivery app (host) and is the exclusive “liaison” of the app.

What it does: Packaging requirements: After you order Kung Pao Chicken, this assistant is responsible for organizing your order (request) according to the standard format of the food delivery platform (MCP protocol).

Delivering orders & waiting for meals: It knows which restaurant’s order taker (MCP server) to send the order to and delivers the order.

Pick up food & report: When the restaurant is ready (execute the operation), it goes to pick up the food (receive the result), then checks it according to the standard format (parse), and finally gives you (the host) the hot Kung Pao Chicken (the final result).

Core function : It is a translator + courier, translating the host's "human language" into MCP's "standard language" and sending it to the server; then translating the server's "standard reply" into "results" that the host can understand and bringing them back.

2.3 MCP Server - "Order Taker + Chef" in the Restaurant Kitchen

Who is it: It is an "intelligent order taker" deployed in places that provide "food" (tools, data, services). It stays next to resources (such as databases, weather APIs, file systems, and smart device control centers).

What it does:

• Order taking: It is always listening for orders (requests). When it receives a standard order (MCP format request) from the assistant (client), it knows what to do (such as "make a serving of Kung Pao Chicken").
• Work: It performs operations. It may cook in the kitchen itself (directly operate resources), or it may instruct the chef in the kitchen (call specific tools or query the database).
• Packaging result: When the dish is ready (the result is out), it will pack the Kung Pao Chicken (result) according to the standard lunch box (MCP standard format) of the food delivery platform.
• Delivery: Finally, it hands the standard lunch box to the assistant (client) waiting at the window.

Core role: It is an executor + packager, understanding standard requests, completing actual work (accessing resources, using tools), and then packaging the results in a standardized manner and waiting for the client to pick them up.

2.4 Summary of the “MCP Takeaway” Process

You (host) say in the food delivery app (AI platform): "I want Kung Pao Chicken!" (initiate a request).

After hearing this, the assistant (client) in the App:

• Put your words into a standard order (packing request).
• Send the order to the order taking counter (MCP server) of XX restaurant.

The restaurant order taker (MCP server) receives a standard order:

• Tell the kitchen staff: "Make some Kung Pao Chicken!" (Perform actions, access resources/tools).
• Once the dish is ready, put it into a standard takeout box (formatted result).
• Hand the takeaway box to the assistant waiting at the window.

The assistant (client) gets the takeaway box:

• Check it out (parse results).
• Run back to the App and hand you (the host) the piping hot Kung Pao Chicken (the final result): "Your order is here!".

Summarize

As I write this, I can't help but recall the time when I was tortured by APIs some time ago. I spent a lot of time looking for documents and testing the API of Feishu Documents and Alibaba Speech Recognition, and it was hard to get them running.

Now, with MCP, these problems can be easily solved by it.

MCP is like a "universal adapter" for AI, allowing it to freely mobilize various tools and services. Whether it is checking the weather, booking air tickets, or controlling smart homes, it is all easy.

As for us developers, we no longer have to stay up all night to modify code in order to connect to the new system.

We look forward to the day when AI truly enters thousands of households with the help of technologies like MCP.