The combination of MCP and blockchain has potential, but it also faces the dual challenges of technical barriers and market pressure.
The significance of AI lies in liberating human labor and raising the lower limit of most work capabilities. However, the current limitations of LLMs are still significant; they require back-and-forth dialogue to provide suggestions, and users must execute the suggestions themselves. We are still some distance away from truly utilizing AI to help us work.
Now, if we could interact with AI to actually use our computers for tasks like replying to emails, writing reports, or even automating cryptocurrency trading, wouldn't we be getting closer to the vision of liberating productivity? This technology is currently the hot keyword in the AI field - MCP.
What is MCP?
MCP (Model Context Protocol) is a set of "standardized agreements" released by Anthropic in November 2024, designed to address the issue of past AI models being able to "speak" but not "do."
First, let's break down the name MCP
Model: Refers to various large AI language models (e.g., GPT, Claude, Gemini, etc.)
Context: Represents additional information or external tools provided to the model.
Protocol: A universal, standardized "specification" or "interface."
Together, it means: through a unified specification, AI can not only "speak" but also directly manipulate external tools to complete various tasks.
Typically, the LLMs we use, such as ChatGPT and Grok, can only perform "text input and text output" based on the dialogue content. If we want AI to help execute operations, such as reading files from a computer folder, sending emails, or querying databases, we usually first give instructions to the LLM, the user then manually operates based on the LLM's response, and finally reports the results back to AI, which then provides us with text suggestions, and we continue the cycle.
The emergence of MCP allows AI not only to read local files on a computer and connect to a remote database but also to directly operate specific online services. In other words, AI is no longer just outputting text; it can complete many repetitive or procedural tasks for you.
Overview of Operation
MCP Host (Administrator): Responsible for managing and coordinating the entire operation of MCP. For example, Claude Desktop is a type of Host that can assist AI in accessing your local data or tools.
MCP Client (User Side): Receives user requests and communicates with the LLM (AI model). Common examples include various chat interfaces or IDEs integrated with MCP (such as Goose, Cursor, Claude Chatbot).
MCP Server: Can be seen as a collection of "organized and annotated" APIs that provide functionalities for AI to use. For example, reading databases, sending emails, managing files, calling external services, etc.
With MCP, AI can not only understand human language but also translate specific text directly into action commands, thereby completing automated operations. For instance, it can help you organize sales reports, send emails to clients, or even perform 3D modeling directly in Blender through commands.
Reference: https://www.youtube.com/watch?v=FDRb03XPiRo&t=4s
Why is MCP Important?
Bridging AI and External Tools
The limitation of LLMs lies in the fact that the data they are trained on is not updated in real-time, meaning the information available to LLMs is limited to what they saw during training. Therefore, any new information generated after training is unknown to the model.
For example, if the LLM was trained in February of this year, it would have no information beyond that date.
The current mainstream method is using RAG (Retrieval-Augmented Generation), which combines a "retrieval system" with a "generation model." This architecture allows for the retrieval of the latest information before LLM reasoning, providing the retrieval results as context to the model. Specifically:
Data Retrieval: Before the LLM answers a question, a retrieval tool (such as web search, internal database queries, etc.) is used to find the latest information relevant to the current question.
Generation: The retrieved data is passed to the LLM as auxiliary information (Context) to help it generate more accurate and timely responses.
For instance, when AI searches for the latest information through Bing or Google before responding to a question and integrates the search results into its response, it is using the RAG method.
The biggest difference between MCP and RAG is:
- RAG uses relatively static data to assist the LLM's responses, while MCP allows AI to truly "take action," such as querying databases, calling APIs, or even modifying file contents.
Standardization & Universality: Like the existence of USB-C: different manufacturers can develop functionalities that comply with MCP specifications, just as all devices can use the same USB-C cable. Without MCP, each developer would have to define how to make AI call specific APIs. This means the same work would be redundantly developed by different people. With MCP standardized, everyone can implement the same set of specifications for immediate integration, avoiding the phenomenon of reinventing the wheel.
From Passive Response to Active Execution: Traditional AI tools only answer questions and cannot take real action. With MCP, AI can decide what commands to execute based on the current situation and read back results to determine the next steps. This ability to continuously adjust based on circumstances significantly enhances the practicality of AI.
Security and Control: MCP does not force all data to be sent to the AI model; it can manage data access through permissions, API key management, etc., ensuring that confidential information does not leak.
What is the difference between MCP and AI Agent?
What is an AI Agent?
In Q3 of last year, GOAT led the trend of AI Agents, and most cryptocurrency users understand AI Agents from a Web 3 perspective. AI Agents typically refer to AI systems that can "automate" specific tasks; they not only converse with humans but also proactively take actions based on context, calling tools or APIs to complete a series of steps. For example, the most common AI Agents can autonomously post on Twitter.
Limitations of AI Agents
Lack of Standardization: Anyone can create an Agent, but without unified specifications, issues arise such as "this Agent only works with model A from vendor A" or "that Agent only calls API B from system B."
Tendency to Operate Independently: Although AI Agents can run errands, developers often need to customize a large number of API formats and rules, leading to a lack of a shared ecosystem between different Agents, making integration difficult.
The relationship between MCP and AI Agents: MCP is a protocol, while AI Agents are a concept or execution method.
AI Agents emphasize the ability of AI to take proactive actions and execute tools.
MCP focuses on how to enable different AI models to communicate with external tools, serving as a universal standard.
MCP helps AI Agents operate more effectively
Without MCP, AI Agents might need to write a separate set of API rules for different tools and platforms, making development and maintenance cumbersome.
With MCP, AI Agents can simply follow MCP specifications to obtain available tools from the "Server list" and dynamically decide which tool to use to complete tasks, making access to external resources safer and more convenient.
Different Functional Scope
AI Agent: Focuses on decision-making and logic, determining how to act and what steps to execute based on needs.
MCP: Specifically addresses tool integration and standard formats, providing a unified way to present external services, databases, and file systems to AI.
Combining the two: AI Agent + MCP = Enabling AI to know how to act and where to act.
What MCP concept projects are currently in the cryptocurrency space?
Base MCP
The framework developed by Base was launched on March 14, allowing AI applications to interact with the Base blockchain. Users can deploy contracts to the blockchain and use Morpho for lending through natural language dialogue without needing development skills.
BORK is the first token deployed using Base MCP, issued on March 14, with a market cap reaching up to $4.6 million, but it has currently fallen back to $110,000, and the 24-hour trading volume is only $90,000, indicating that the token's lifespan has likely ended.
Flock is a decentralized AI training platform that points out that the current MCP still operates on external AI models for centralized LLM processing. Flock provides a Web3 proxy model, allowing AI-driven blockchain tasks to run locally, thereby providing users with more control.
Lyra
LYRAOS, short for LYRA MCP-OS, is a multi-AI Agent operating system that allows AI Agents to interact directly with the Solana blockchain to perform operations such as buying and selling cryptocurrencies.
They are currently exploring how to use MCP-OS to establish thousands of "AI16ZDAOs," which are AI-driven decentralized autonomous organizations for cryptocurrency investment. LYRAOS plans to release a DEMO between March 21 and 22, 2025, and launch the official product next week.
Current token market cap is $923,000, with a peak of $2.64 million, 24-hour trading volume of $3 million, and 2,922 token holder addresses.
Conclusion: The AI narrative is dancing again, but it will take time to observe.
Although MCP provides a standardized rule that allows AI to interact more easily and safely with external tools, and seems to have great potential in the Web 3 space, successful cases are relatively limited. The reasons behind this may include the following points:
Technical integration is not yet mature: In the Web 3 ecosystem, each chain and each DApp has different contract logic and data structures. Unifying them into an MCP Server that can be called by AI still requires a significant investment of development resources.
Security and regulatory risks: Allowing AI to directly manipulate contracts and handle financial transactions requires a well-designed private key management and permission control mechanism, which is both difficult and costly.
User habits and experience: Most people are still skeptical about letting AI manage wallets or make investment decisions, and the operational threshold of blockchain itself is high. If the experience is too complex or lacks clear application scenarios, beginners will find it difficult to use or invest in the long term.
Aesthetic fatigue and market indifference: Previously, AI Agents created a wave in the cryptocurrency space, with many unimplemented projects reaching valuations of over a hundred million during their peak. Recently, however, we are facing a phase of bursting the AI bubble, with most projects dropping over 90%, which is seen as a disillusionment with AI.
Returning to the MCP narrative, it can be understood as a super-enhanced version of AI Agents. The market has already experienced a cryptocurrency AI frenzy and gradually understands the difference between concept hype and practical application. If there are no truly innovative and practical applications, investors and users will not easily buy in. Early MCP projects like BORK did not generate significant hype due to a lack of clear differentiation or practical application, which is a key factor in why the current MCP concept has not gained widespread popularity.
The combination of MCP and blockchain has potential, but it also faces the dual challenges of technical barriers and market pressure. In the future, if more mature security mechanisms can be integrated, a more intuitive user experience can be created, and truly valuable innovative applications can be discovered, "Web 3 + MCP" may break free from the fate of being merely a "hyped topic" and become a new main narrative.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
