Author: CG Zhou Yijian Source: X, @Cyrus_G3
After Agent, MCP has become the biggest hit in the AI track in 2025. This open source protocol developed by Anthropic is committed to becoming the HTTP protocol in the AI world. It has been integrated by large AI services such as Microsoft; MCP concept currency #DARK has risen against the trend recently. FDV has risen from 10 million US dollars in early April to 30 million US dollars. So, what is MCP? What are the points of convergence between MCP and Crypto projects? This issue of "CG Vernacular Science" will use an article to help you thoroughly understand MCP.
1. What is MCP?
MCP, the full name of Model Context Protocol, is an open standard protocol in the field of AI. You can think of it as a "universal plug" in the AI world. Just like a universal plug that allows various electrical appliances to be plugged into the same socket, MCP allows AI models to easily connect to various external tools and data sources, such as databases, file systems, or third-party APIs. It is an open source project developed by Anthropic, specifically to solve the trouble of connecting AI with external resources.
It should be noted that MCP is mainly used for applications that use AI models (such as AI assistant software), rather than directly for the AI model itself. Simply put, it helps AI applications to "bridge" with the outside world more conveniently.
2. What problems does MCP solve?
Before the emergence of MCP, it was not so easy to let AI "chat" or "connect" with external tools or data. Developers had to write separate code for each data source or tool, just like every time you bought an electronic product, you had to buy a dedicated charger, which was time-consuming and laborious. For example, if you want AI to check something from the database or organize the files in the computer, you have to write the code from scratch every time, which is too troublesome. MCP is like a "universal charger" that provides a standardized connection method. As long as the AI application supports MCP, it can easily connect to various external resources, saving time and effort and improving efficiency. To put it in a more vivid way, MCP is like USB Type-C, which is used by most of our mobile phones now, which is specifically designed to help AI applications easily connect to external data. If the MCP protocol is widely promoted, it will greatly increase the development and popularization speed of AI applications.
3. How does MCP work?
MCP operates in a "client-server" mode, which sounds high-end, but is actually very simple:
MCP host: It is actually the AI application you use, such as a software called Claude Desktop. It is the "client" responsible for greeting external resources.
MCP Server: It is actually a small service program that connects to a specific data source or tool, such as your file system or an API.
Communication method: They use standardized "languages", such as JSON-RPC or gRPC protocols, to ensure safe and smooth communication.
For example: Suppose you want an AI assistant to help you organize files on your computer. Through MCP, your AI application (host) can be directly connected to the file system (server), and then AI can read and write files and manage folders without having to write a lot of code for each file type. Just like you can control the lights and air conditioners at home with a remote control, it is very convenient.
IV. Current problems of MCP
Although MCP sounds great, it still has some problems waiting to be solved:
1. Security: When dealing with external data, especially sensitive information (such as your encrypted wallet account), security is a big problem. MCP must ensure that this data will not be abused or leaked. The @darkresearchai project uses TEE (Trusted Execution Environment) to protect the security of encrypted accounts, so that Agent can safely use the MCP protocol to complete a series of on-chain operations.
2. Universality: At present, MCP has not yet been truly popularized. At present, it is mainly integrated with Microsoft AI services, and many large AI services have not yet integrated MCP. For MCP to really become powerful, more developers and companies must join in to support it. Now MCP is still a "newcomer", many people have not started using it, and the ecosystem is not large enough. This has led to many AI applications developed based on large models being unable to use MCP directly.
3. Compatibility: MCP must ensure that it can be used on different systems and platforms. If it cannot be used in some places, it will be very embarrassing. Just like when smart homes first came out, everyone was worried about whether they would be hacked or some devices could not be connected at all. MCP now has similar problems.
V. The future value of MCP
If MCP can be widely accepted, the future will be unlimited. It may become the "universal language" for AI to communicate with external resources, just like the Internet uses HTTP. At that time, AI applications will become smarter, more capable, and can do more things.
For example, your AI assistant can directly dig data from the company database, help you manage your schedule, and even work seamlessly with other services, without you having to worry about the whole process. At that time, the real DeFi Agent will be established, and the Agent can traverse the entire chain to help you manage the assets on the chain and help you capture the on-chain income safely and efficiently. This is the real DeFAI.
Imagine that if MCP becomes a standard, developers will not have to write connection code "from scratch" every time, and the development speed of AI applications will be faster and the cost will be lower. In the end, ordinary people like us can also use more powerful and convenient AI tools. MCP is still in the development stage, and its future potential has not been fully realized, but the prospects are definitely worth looking forward to.
Summary
In general, MCP is a "big move" in the field of AI. It uses a standardized protocol to make AI applications easily connect to various tools and data sources, solving the tedious problems in previous development. Although there are still challenges such as security, adoption rate and compatibility, if they can be overcome, MCP may completely change the way AI applications are played, making them more powerful and more popular. Just like the evolution of mobile phones from "brick phones" to smart phones, MCP may be the key driver for AI to move to the next step! By then, the combination of Crypto and AI will become more convenient and efficient. I believe that there will be a series of MCP-based on-chain Agent applications worth looking forward to.
Kikyo
Clement
Catherine
Joy
Davin
Hui Xin