The problem The MCP Agent Hub for Web3 solves

On-chain AI agents are quickly gaining traction across blockchains like Ethereum, NEAR, Rootstock, Saga, and others. But with each chain using different API formats, transaction models, and discovery mechanisms, the ecosystem is becoming increasingly fragmented,making it hard for developers and users to keep up.

First, there’s the discovery problem: there’s no easy way to find the right AI agents for specific tasks like DeFi trading, DAO governance, or IP licensing. Without a shared index or reputation system, users have to hunt through individual repos, documentation, or community threads just to figure out what’s available and trustworthy.

Then comes integration complexity. Even after identifying a useful agent, getting it to work means writing custom code for every chain, dealing with inconsistent authentication flows, and handling payments in ways that vary from agent to agent. It’s tedious and time-consuming.

Monetization adds another layer of messiness. Some agents require prepaid tokens, others expect escrow, and many don’t support streamlined payments at all. The user experience is inconsistent, and for developers, there’s no standard way to earn income from their agents.

Finally, there’s the trust issue. With new agents popping up all the time, it’s tough to tell which ones are actually reliable. There’s little visibility into how an agent performs, how often it’s used, or whether others have had success with it.

The MCP Agent Registry is built to solve these problems. It creates a common layer where AI agents can register, describe their capabilities, and handle payments, all in a standardized way. Users get a single portal to browse agents, see usage stats and ratings, and call agents via the Model Context Protocol. Developers, in turn, get a unified platform for distribution, monetization, and cross-chain compatibility.

Challenges we ran into

One major challenge was making the system work across multiple blockchains like Ethereum, NEAR, Rootstock, and Saga. Each has its own RPC endpoints, signing methods, and transaction rules. To unify them under a single protocol, we built custom MCP bridge nodes for each chain. These nodes handle the chain-specific details behind the scenes so that developers can interact with any agent in the same way.

Payments were another complex area. Since agents might charge in different tokens, ETH, NEAR, rBTC, we needed a standard contract layer to collect fees consistently. Bridging tokens and normalizing payment flows across chains took careful planning and rigorous testing to get right.

Helping users find the right agent also required building a ranking system. We trained a custom preranker model on crypto-specific queries and agent metadata to surface relevant results. High-quality training data was hard to find, so we gathered logs from real agent interactions and curated domain-specific text from areas like DeFi and DAO governance. Fine-tuning the model took several iterations, but it now accurately matches user needs to the right agent.

Some agents rely on compute-heavy tasks like parallel ML inference or analytics. To support that, we implemented an off-chain job orchestrator with a proof-of-completion mechanism on-chain. Ensuring the system stayed reliable even with node failures or network lag was a significant infrastructure challenge.

Finally, packaging all this into a usable interface meant we had to balance technical depth with clarity. We added real-time usage stats, success rates, and community feedback to help users trust what they’re seeing. That required robust logging and privacy-conscious design both on-chain and off-chain.