In November last year, Anthropic released the MCP protocol MBTI professional test . At this time, the Sentinel team included it in the list of key tracks. Nowadays, as MCP-based Agent products such as Manus begin to become popular, this relatively niche technical term is stepping out of the circle, and some people even claim that it may become CUDA in the AI era. Just today, we are going to use vernacular to clearly explain what MCP is, and also explain the similarities and differences between it and CUDA.
The stuck problem developed by Agent
In 2023, OpenAI launched the function calling function. Since then, developers finally have the ability to make AI models call external tools. An engineer from an AI startup in San Francisco told me that this is indeed a milestone. However, the trouble arises: the days of writing code for each tool have begun.
This company that engages in AI mail processing complained during an interview that developers write specific business logic for each system and lack a unified interface standard. This means that most of the energy is spent on integration details instead of innovation in core AI functions. This fragmentation seriously slows down the progress of Agent development.
One universal adapter does everything
In order to solve this pain point, Anthropic launched the model context protocol. You can think of it as a universal adapter that allows the AI model to communicate with different tools using a unified language without having to write special code for each tool, just like the USB-C interface unifies the charging cable.
To put it simply, MCP defines a set of standard guidelines, which clarify how the AI model calls the external tool MBTI personality test , how to obtain data, and how to interact with the service. In this way, developers only need to access it once, and AI can connect various data sources and tools on its own, significantly reducing the complexity of development.
From code completion to autonomous workflow
The design inspiration of MCP actually comes from the LSP protocol in the programming field. It is LSP that enables the code editor to communicate with the language server to achieve the automatic completion function. However, MCP goes one step further on this basis. It supports AI to make autonomous decisions, rather than passively responding to requests.
This shows that the AI Agent can determine on its own which tools to use the MBTI test and in what order to use them, and even allows humans to participate to provide data or approve execution. At the end of last year, a developer in Silicon Valley excitedly said on Twitter that this had finally solved the interface problem for human-machine collaboration.
Thousands of servers have sprung up in the community
When it was released in November 2024, MCP did not cause much fluctuation. However, by the beginning of 2025, as Agent became a hot topic, the situation changed dramatically. According to official February data, the community has built more than 1,000 MCP servers.
Companies such as Block and Replit have integrated it into the system, and development tools such as Zed and Sourcegraph have also begun to rely on MCP to strengthen the platform. The Manus application that hit the screen not long ago is a very typical example of MCP plus cloud environment, which made many people see the power of MCP for the first time.
Is it similar to CUDA?
Nowadays, many people compare MCP to CUDA. The two do have similarities, that is, they both simplify the development difficulty of their respective ecosystems and solve key needs. However, the essential difference is that CUDA is closely related to NVIDIA hardware, while MCP is an open source and model-independent protocol that can be used by any AI.
CUDA is mainly used to accelerate model training and inference. It already has nearly two decades of accumulation. MCP was released only a few months ago. It mainly solves the problem of connecting external tools to AI. However, with the advantages of open source and the promotion of Anthropic, it is quickly transforming into a de facto standard in Agent development.
Open protocols hide business ambitions
Although the current MCP is not bound to a specific model and cannot form a vendor lock-in situation similar to CUDA in the short term, if it becomes a widely used standard, it can still bring great advantages to Anthropic. Combined with Claude's leading programming capabilities, they are offloading developers from OpenAI.
Anthropic has planned to provide a toolkit for deploying remote production MCP servers, and the community is also actively promoting remote server support. This shows that more products such as Manus will appear in the future, and the Agent ecosystem will usher in explosive growth. Do you think MCP will eventually unify the AI tool interface like USB-C, or will it be replaced by the closed standards of each major manufacturer? Welcome to share your views in the comment area, like and forward to let more people know about this new AI trend.
