AI Agent Observation Series Issue 3

After the AI Agent track demystified MemeCoins, it began to explore new development directions. Recently, the MCP and A2A protocols have attracted market attention; both initially developed in the Web2 space and have recently been introduced to Web3. MCP (Model Context Protocol) is an open protocol co-built by the community and launched by Anthropic. The introduction of the MCP protocol is expected to link LLM with external "knowledge" and "actions," providing standardized perception and execution interfaces for AI Agents, similar to a universal interface for AI Agents. A2A (Agent-to-Agent), launched by Google, aims to promote collaboration between different AI Agents. The introduction of the A2A protocol is expected to facilitate autonomous communication and collaboration between AI Agents, enabling them to automatically complete on-chain dynamics.

In the context of the new protocols, a number of new exploratory projects have emerged, along with the evolution of existing AI Agents represented by arc under the new protocols. However, the current situation shows that the market is cautious about new narratives. Most projects are exploring based on the MCP protocol, and there are also projects that combine the concepts of MCP and A2A protocols. However, without actual products, the enthusiasm for such projects can usually only last a few days, with a market cap exceeding $10 million considered outstanding. Overall, the market under the new protocol context is not optimistic.

Total Market Cap of AI Agents Rises by 18.6%

According to data from cookie.fun, as of the time of publication, the total market cap of AI Agents is $7.33 billion. Notably, from the second phase of the AI Agent observation series (February) analyzed by CoinW Research Institute to the current third phase (with a total market cap of $6.18 billion), the total market cap of AI Agents has increased by 18.6%.

Currently, there are a total of 1,589 AI Agents. Compared to the main distribution of AI Agents on the Solana and Base chains in February, the current main public chains for AI Agents are Solana and Ethereum. Therefore, it can be observed that Solana, relying on its unique advantages, remains the primary public chain for AI Agents. Additionally, the Ethereum chain is gradually beginning to take on new traffic from AI Agents.

_Figure 1 Source: _cookie.fun

Currently, the total market cap of AI Agents on the Solana chain is $2.67 billion, accounting for 36.43%; the total market cap of AI Agents on the Ethereum chain is $2.41 billion, accounting for 32.89%. According to cookie.fun data, the top three AI Agents by market cap are $VIRTUAL with a market cap of $870 million, $AI16Z with a market cap of $300 million, and $AIXBT with a market cap of $140 million. Below, CoinW Research Institute will filter out relatively outstanding projects in this track based on comprehensive dimensions.

Note: Most AI Agent projects are still in the early stages and are considered high-risk investments, with significant token volatility. This article is for information sharing and discussion only, not investment advice, DYOR.

Recent AI Agents Worth Noting

arc

arc once became a leading framework project in the AI Agent track with its unique RIG framework. Following market developments, arc recently announced the upcoming launch of the agent-based application store Ryzome, which is deeply integrated with the MCP protocol. arc believes that future AI Agents should no longer operate in isolation but should collaborate and evolve with each other. Thus, Ryzome was born.

Ryzome unifies service mechanisms and standardized interfaces through the MCP architecture, simplifying the process for AI Agents to integrate third-party tools and effectively reducing the development and maintenance costs associated with tool fragmentation. Its modular and composable design enhances system scalability, allowing agents to efficiently adapt to diverse task scenarios. Additionally, Ryzome leverages the native infrastructure provided by Anthropic to further optimize the developer experience in terms of high availability and standardized access processes, improving system stability and development efficiency.

_Figure 2 Source: _arc.fun/ryzome-waitlist

Currently, Ryzome is still in a closed internal testing phase, and users can apply for access through a whitelist. After Ryzome officially launches, users will access it through interfaces like Claude Desktop, cline, and cursor that comply with MCP standards. Notably, after Ryzome officially launches, arc will also introduce a product called Canvas. The two are a series of products, with Ryzome primarily providing outreach capabilities for agents, while Canvas mainly facilitates organization and coordination among agents. Although the products have not yet launched, it is clear that arc is continuously evolving its underlying system. As of the time of publication, the market cap of arc's native token is $67.08 million.

SkyAI

SKYAI is based on the BNB Chain and has received support from the leading launch platform on BNB Chain, Four.meme. The core advantages of SKYAI are mainly reflected in the following points: achieving multi-chain data interoperability through the MCP protocol, breaking the data silo problem; providing standardized development tools and pre-built AI modules, significantly lowering development barriers; and forming a complete end-to-end application loop around natural language interaction, intelligent analysis, and risk protection.

SkyAI is also the first pre-sale project of Four.meme. According to BSCDaily data, a total of 112,306 users participated in the SkyAI pre-sale, raising a total of 83,848 BNB, over-subscribed by about 167 times. It is worth noting that the SkyAI project began operations on April 11, 2025, and currently does not have any actual products launched. From the official website, the project’s site is very simple, and the roadmap is only a rough plan. As of the time of publication, the market cap of SkyAI's native token $SKYAI is $41.49 million.

_Figure 3 Source: _skyai.pro

DARK

DARK is an experimental MCP network based on the Solana chain, operating in a Trusted Execution Environment (TEE). It received widespread attention at its launch, and its developer (DEV) Edgar Pavlovsky is the organizer of the next hackathon under the Solana official organization. Edgar Pavlovsky is also the founder of mtndao (the core hacker community of Solana). He has participated in the development of Marginfi (a decentralized finance protocol) and has rich development experience and relevant VC incubation background. Edgar Pavlovsky positions DARK as an infrastructure combining AI and GameFi, thus DARK possesses the attributes of Meme propagation while also having a long-term narrative as a foundational protocol.

Currently, DARK's first product is still in development, and users can join the waiting list by registering their email. At the same time, DARK is also developing an AI competitive game called DARK Games, where players need to use the native token $DARK to participate. Additionally, it is important to note that DARK's official GitHub repository has only submitted 2 contents. As of the time of publication, the market cap of $DARK reached a high of $50 million but has now fallen to $24.69 million.

_Figure 4 Source: _github.com/orgs/darkresearch

OpenServ

OpenServ, through integration with MCP, allows AI agents to seamlessly access over 10,000 tools. OpenServ facilitates the transition of AI Agents from experimental tools to practical tools. Like other projects, OpenServ is also in the early stages. Currently, the OpenServ builder experimental field is only open by application, and users can experience it only after passing the review. Additionally, from the GitHub code repository, OpenServ began submitting relevant code in March, and currently, only 6 repository contents have been submitted.

In April of this year, OpenServ held the DeFAI hackathon, attracting over 1,500 registrants and more than 100 innovative agents. Currently, OpenServ has integrated with Telegram, and in future plans, it will also deeply integrate with AWS, Notion, Figma, Google Maps, and more. From the community sentiment, OpenServ is clearly lagging behind other projects. As of the time of publication, the market cap of OpenServ's native token $SERV is $28.93 million.

Summary

Protocols like MCP are undoubtedly the latest highlights for AI Agents, with new projects like SkyAI emerging and existing AI Agent projects evolving. Additionally, projects like Cookie DAO and Holoworld, which focus on DEFAI, are gradually shifting towards exploring MCP and A2A. However, from the current project situation, protocols like MCP that have transitioned from Web2 do not naturally adapt to Web3.

This article believes that in this context, to truly help the AI Agent track break out of the narrative trap, it may be worth considering and optimizing from the following two angles. First, consider the reconstruction of protocol layer adaptability. Project parties can reconstruct the protocol layer from a Web3 native perspective, introducing composable on-chain behavior standards for agents to address specific on-chain issues such as data alignment and event synchronization between chains; secondly, consider linking the token model with the lifecycle of agents. The economic mechanism design of the project can be directly tied to the operational status of agents, such as reward distribution logic based on agent call frequency, outcome verification, and on-chain scoring mechanisms, activating the real utility of tokens, driving product landing and interaction, thereby promoting the long-term development of the project.

Overall, the AI Agent track is still waiting for a new breakthrough.

Reference Links: