Q&A: What is the MCP protocol? If you don’t understand MCP, will you miss out on a billion?

MCP protocol: the "USB" of large language models, connecting AI to the outside world!
Core content:
1. The definition and function of the MCP protocol: the "USB" protocol of large language models
2. How does MCP allow large language models to call external services: taking weather forecasts and querying chat history as examples
3. How can developers seize the MCP ecological niche and provide valuable services to users
Q: Teacher Baoyu, please tell us more about MCP when you have time. I really don’t have time to keep up with the information during this period of time. I feel that I have fallen behind a lot. I feel like I have missed a billion every day . A: MCP is currently more of a toy for developers. Ordinary users do not need to pay special attention to it. When the ecosystem is established and the configuration is simple, just use the ready-made good MCP service. The current limitation of the large language model is that it cannot directly connect to the external real environment. For example, you don’t know what day it is today, what the weather is like outside, you can’t use a calculator, and you can’t search or access the web. It’s like you have a computer, but there is nothing but a keyboard, mouse, and monitor. You can’t access the Internet, and you can’t connect to a scanner, camera, etc. MCP is a protocol (MCP: model context protocol) used to regulate the communication between large language models and external tools. Just like the USB protocol on your computer, with the USB protocol, your computer can connect to various external devices, such as a camera with a USB interface, which can record videos; a scanner with a USB interface, which can scan documents; and a USB network card, which can surf the Internet. MCP is the USB protocol of the large language model. Developers can develop various services based on this protocol, such as searching the web, connecting to your local WeChat chat history, weather forecast services, etc.
So how does the large language model know which service to call? Here you need to have an operation to register a service. Now it is a JSON configuration file, which configures each service in it, just like your computer has a USB Hub, and a bunch of your devices are plugged into the USB Hub, and the computer can sense which devices are plugged into the USB Hub. Of course, the HUB of the large language model is smarter, and you don’t need to choose which device (service) you want. The model will select the appropriate service from the registered services according to your requirements. For example, your chat application has registered several MCP services: weather forecast, network search, query local WeChat chat history, etc. You ask ChatBot: What will the weather be like in Shanghai in the next three days? ChatBot knows that it can’t answer this question, but it knows that there is a weather service in the registered MCP service, so it converts the user’s question into parameters: date range (03/18/2025-03/20/2025), city: Shanghai, and then calls the weather service. The weather service returns the weather results of these three days to the model, and then the model can reply to the user after getting the results: Today and tomorrow will be sunny, suitable for outdoor; the day after tomorrow will rain, remember to bring an umbrella. Then you ask ChatBot: Give me a summary of yesterday's chat history of XX group. ChatBot does not have your chat history, but it can look for WeChat chat history to query the MCP service, pass in the parameter (date: 03/17/2025), and the service returns yesterday's chat history to ChatBot, which summarizes and tells the user the result. If you are a developer, you can first make some simple applications to understand what it is. If you think you can provide valuable services, you can also seize the ecological niche first, make some applications, and consider how to access payment. You can make money by selling MCP services in the future. There are still some opportunities in the future. For users, the current MCP is still in a relatively elementary state. It is not easy to use, and it will not immediately change your user experience. Keep understanding and wait and see. If there is an MCP application scenario that can really improve your daily work efficiency, then you must try it. Maybe if you use it earlier, you can greatly improve your efficiency, and you don’t need to do any development work yourself. It’s like buying a new USB device for your computer, which provides you with a lot of convenience.