New Explorations of AI Agents in the Web3 Domain: From Manus to MCP

Recently, a product named Manus, the world’s first universal AI Agent, has attracted widespread attention. This product, developed by a Chinese startup, has created a stir in the tech circle, showcasing the immense potential of AI Agent technology with its powerful independent thinking and execution capabilities. The success of Manus has not only drawn industry attention but has also provided valuable design ideas for the development of AI Agents, with various fields, including the Web3 industry, exploring its application prospects.

An AI Agent is an intelligent system capable of making decisions and executing tasks autonomously based on its environment, inputs, and predefined goals. Its core components include a large language model (LLM) as the “brain”, as well as key functions like observation, reasoning, execution, and memory. Currently, there are two main development paths for AI Agent design patterns: one focusing on planning capabilities and the other on reflective capabilities. Among them, the ReAct pattern is the earliest and most widely used design pattern, with its core process being a cycle of thinking, acting, and observing.

In the Web3 field, although the popularity of AI Agents has somewhat diminished, there are still some projects actively exploring. These mainly include three models: Launch platform model (such as Virtuals Protocol), DAO model (such as ElizaOS), and commercial company model (such as Swarms). Among them, the launch platform model is currently more mature and able to achieve an economic closed loop. However, this model also faces challenges due to insufficient intrinsic value support.

The emergence of the Model Context Protocol (MCP) has brought new exploration directions for AI Agents in Web3. MCP is an open-source protocol designed to solve the connection problem between LLMs and external data sources. In the field of Web3, the application of MCP mainly has two directions: first, deploying the MCP Server on blockchain networks to achieve decentralization and censorship resistance; second, empowering the MCP Server with the capability to interact with blockchains, thereby lowering the technical threshold. Additionally, some scholars have proposed a plan to build an OpenMCP.Network creator incentive network based on Ethereum.

Although the combination of MCP and Web3 theoretically can inject decentralized trust mechanisms and economic incentives into AI Agents, it still faces challenges in technical validation and efficiency issues. For Web3, the integration with AI is an inevitable trend that requires continuous exploration and innovation in the industry.

AI, as a grand technological revolution, will undoubtedly have a profound impact on Web3. Although there are still many challenges at present, with continuous technological advancements and the emergence of innovative ideas, the application prospects of AI Agents in the Web3 field remain broad. We look forward to seeing more groundbreaking products and solutions that promote the deep integration of Web3 and AI, creating more value.