Are AI Agent Tokens Falling Because MCP Is Too Hot?
TechFlow Selected TechFlow Selected
Are AI Agent Tokens Falling Because MCP Is Too Hot?
The key challenge for a true web3 AI Agent lies in aligning the AI Agent's "complex workflow" with blockchain's "trust verification flow" as seamlessly as possible.
Navigating Web3 tides with focused insightsAuthor: Haotian
Some friends have suggested that the continuous decline of web3 AI Agent projects like #ai16z and arc is caused by the recent surge in popularity of the MCP protocol? At first glance, I was completely confused—WTF, is there even a connection? But upon deeper reflection, there actually is some logic here: the valuation framework for existing web3 AI Agents has changed, and their narrative direction and product roadmap urgently need adjustment! Below are my personal thoughts:
1) MCP (Model Context Protocol) is an open-source standardized protocol designed to seamlessly connect various AI LLMs/Agents to diverse data sources and tools—analogous to a plug-and-play USB "universal" interface that replaces the previous end-to-end "customized" integration approach.
In simple terms, AI applications used to operate within clear data silos. For Agents/LLMs to exchange information, each had to develop specific API interfaces—a complex process lacking bidirectional interaction, often limited in model access and permission controls.
The emergence of MCP provides a unified framework enabling AI applications to break free from these data silos, making it possible to "dynamically" access external data and tools. This significantly reduces development complexity and improves integration efficiency, especially accelerating automation tasks, real-time data queries, and cross-platform collaboration.
At this point, many might immediately think: if Manus—the innovator in multi-Agent collaboration—were to integrate this MCP open framework that enhances multi-Agent cooperation, wouldn't that be unstoppable?
Exactly. It's precisely the combination of Manus + MCP that constitutes the core reason behind the current market pressure on web3 AI Agents.
2) However, curiously enough, both Manus and MCP are frameworks and protocol standards aimed at web2 LLMs/Agents. They address data interaction and collaboration between centralized servers, relying on individual server nodes to "actively" open up permissions and access. In other words, they remain purely open-source tooling.
In theory, this runs completely counter to the core principles pursued by web3 AI Agents—"distributed servers, distributed collaboration, distributed incentives." How can a centralized cannon destroy a decentralized fortress?
The root cause lies in how overly "web2-like" first-generation web3 AI Agents have been. Many teams originate from web2 backgrounds and lack sufficient understanding of native web3 needs. For example, ElizaOS started as a wrapper framework helping developers quickly deploy AI Agent applications—essentially integrating platforms like Twitter, Discord, and APIs from OpenAI, Claude, DeepSeek, while providing generic wrappers for Memory and Character modules to speed up AI Agent development. But strictly speaking, what differentiates this service framework from web2 open-source tools? What unique competitive advantages does it offer?
Uh… Is the only advantage a Tokenomics incentive model? Then using a framework that web2 could easily replicate to incentivize a batch of new tokens solely for token issuance? Scary... Following this logic, you begin to see why Manus + MCP could disrupt web3 AI Agents.
Since most web3 AI Agent frameworks and services only solve the same rapid development and deployment needs as their web2 counterparts—but fail to keep pace with web2 innovation in technical standards and differentiation—markets and capital have begun re-evaluating and repricing last-generation web3 AI Agents.
3) By now, we’ve likely identified the root issue. So how do we break through? There’s only one path: focus exclusively on building web3-native solutions, because distributed system operations and incentive architectures are where web3 holds absolute, differentiated advantages.
Take distributed cloud computing power, data, and algorithmic service platforms as examples. On the surface, aggregating compute and data resources based on idle capacity seems insufficient in the short term to support engineering-level innovation. While mainstream AI LLMs are engaged in centralized computational arms races chasing performance breakthroughs, a service model based on "idle resources" and "low cost" naturally appears unattractive to web2 developers and VC firms.
But once web2 AI Agents move beyond the phase of pure performance competition, they will inevitably shift toward vertical application expansion and fine-tuning specialized models—this is when web3 AI resource services will truly shine.
In fact, when web2 AI players reach a certain stage via resource monopolization, it becomes nearly impossible for them to revert to a "rural encirclement of cities" strategy, conquering one niche scenario at a time. That moment will mark the rise of collaboration between surplus web2 AI developers and web3 AI resources.
Indeed, beyond the web2 playbook of rapid deployment + multi-Agent communication frameworks + Tokenomic token narratives, there are numerous web3-native innovation directions worth exploring for web3 AI Agents:
For instance, implementing a distributed consensus collaboration framework, considering the nature of off-chain computation for large LLMs coupled with on-chain state storage, which requires multiple adaptive components:
1. A decentralized DID identity verification system allowing Agents to possess verifiable on-chain identities—similar to how virtual machines generate unique addresses for smart contracts—primarily enabling continuous tracking and recording of states;
2. A decentralized Oracle system responsible for trustworthy acquisition and validation of off-chain data. Unlike traditional Oracles, this AI Agent-adapted Oracle may require a composite architecture involving multiple Agents across data collection, decision consensus, and execution feedback layers, ensuring real-time delivery of required data, off-chain computation, and decision-making to on-chain systems;
3. A decentralized DA (Data Availability) storage system. Given the uncertainty of knowledge base states during AI Agent operation and the ephemeral nature of inference processes, a system is needed to record and store key backend LLM state databases and inference paths within a distributed storage network, offering cost-effective data availability proofs to ensure data accessibility during public chain verification;
4. A zero-knowledge proof (ZKP) privacy computation layer integrating privacy-preserving solutions such as TEE and FHE, enabling real-time private computation and data proof verification. This allows Agents to access broader vertical data sources (e.g., healthcare, finance), paving the way for more specialized, customized service Agents;
5. A cross-chain interoperability protocol—somewhat similar to the framework defined by the MCP open protocol—but differing in that this Interoperability solution must include relays and communication scheduling mechanisms tailored for Agent execution, transmission, and validation, capable of handling asset transfers and state synchronization across chains, particularly including complex states such as Agent context, prompts, knowledge bases, and memory;
...
In my view, the true challenge for web3 AI Agents lies in aligning the "complex workflow" of AI Agents with the "trust verification flow" of blockchains as closely as possible. Whether these incremental solutions emerge from upgrades and iterations of existing legacy narrative projects or are forged anew by emerging projects within the AI Agent narrative space—both possibilities exist.
This is the direction web3 AI Agents should strive to build toward—the foundational innovation ecosystem aligned with the grand macro narrative of AI + Crypto. Without meaningful innovation and the establishment of differentiated competitive moats, every ripple in the web2 AI sector could send shockwaves through the entire web3 AI landscape.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
Haotian | CryptoInsight
