The Exploration of AI Agents in the Web3 Field: From Manus to MCP

Recently, a product called Manus, the world's first general-purpose AI Agent, has sparked heated discussions in the domestic tech circle. Developed by a Chinese startup, the product saw overwhelming demand for invitation codes on its launch day. As a general-purpose AI Agent, Manus demonstrates powerful capabilities in independent thinking, planning, and executing complex tasks, able to autonomously complete the entire process from conception to delivery.

The explosive popularity of Manus has not only attracted attention in the industry but also provided valuable product ideas for AI Agent development. With the rapid development of AI technology, AI Agents, as an important branch of artificial intelligence, are gradually moving from concept to practical application, demonstrating enormous potential across various industries, and the Web3 field is no exception.

An AI Agent is a computer program that can autonomously make decisions and perform tasks based on the environment, input, and predefined goals. Its core components include a large language model (LLM) as the "brain", observation and perception mechanisms, reasoning and thinking processes, action execution capabilities, as well as memory and retrieval functions.

The design patterns of AI Agents mainly have two development routes: one emphasizes planning capabilities, including REWOO, Plan & Execute, LLM Compiler, etc.; the other emphasizes reflective abilities, including Basic Reflection, Reflexion, Self Discover, LATS, etc. Among them, the ReAct model is the earliest and most widely used design pattern.

The ReAct mode addresses diverse language reasoning and decision-making tasks by combining reasoning and acting in language models. Its typical process can be described by the "Think→Act→Observe" (TAO) cycle.

The AI Agent can also be divided into Single Agent and Multi Agent based on the number of agents. Single Agent focuses on the combination of LLM and tools, while Multi Agent assigns different roles to different agents, completing complex tasks through collaboration.

Model Context Protocol (MCP) is an open-source protocol launched by Anthropic, aimed at solving the connection and interaction issues between LLMs and external data sources. MCP provides three capabilities: knowledge extension, function execution, and pre-written prompt templates, using a Client-Server architecture and employing the JSON-RPC protocol underneath.

In the Web3 field, the development of AI Agents has experienced peaks and troughs. Currently, there are three main models: the launch platform model represented by Virtuals Protocol, the DAO model represented by ElizaOS, and the commercial company model represented by Swarms.

The launch platform allows users to create, deploy, and monetize AI Agents. Virtuals Protocol is currently the largest launch platform, with over 100,000 Agents issued on it. The DAO model, such as ElizaOS, aims to utilize AI models to simulate investment decisions and make investments based on suggestions from DAO members. Swarms is an enterprise-level Multi-Agent framework that addresses complex business needs through intelligent orchestration and efficient collaboration.

From the perspective of economic models, currently only launch platforms can achieve a self-sustaining economic closed loop. However, this model also faces challenges, mainly because the issued AI Agents mostly lack intrinsic value support.

The emergence of MCP has brought new exploration directions for Web3's AI Agent: first, deploying MCP Server to the blockchain network to solve single point issues and have censorship resistance; second, enabling MCP Server to interact with the blockchain, such as conducting DeFi transactions and management.

In addition, there is a proposal for a creator incentive network called OpenMCP.Network built on Ethereum. This proposal aims to achieve automation, transparency, trustworthiness, and censorship resistance of incentives through smart contracts, and utilizes technologies such as Ethereum wallets and ZK for signature, authorization verification, and privacy protection during the operation process.

Although the integration of MCP with Web3 theoretically injects decentralized trust mechanisms and economic incentive layers into AI Agent applications, the current zero-knowledge proof technology still struggles to verify the authenticity of Agent behavior, and decentralized networks still face efficiency issues. This is not a solution that can succeed in the short term.

The release of Manus marks an important milestone for universal AI Agent products. The Web3 world also needs a milestone product to break the external doubts about its lack of practicality and being just hype. The emergence of MCP brings new exploration directions for Web3's AI Agent, including deployment on blockchain networks, achieving interaction with blockchains, and building creator incentive networks.

AI, as a grand historical narrative, is inevitably merging with Web3. We need to maintain patience and confidence while continuously exploring the development of this field.