Building the Future of AI and Web3 with MCP

Introduction

This article aims to explore the architecture design, technical decisions, and engineering strategies that supported the growth of REVOX MCP from the perspective of the REVOX technical team.

Phase 1: Core Infrastructure Development

The focus of the first phase was to build the core infrastructure of the AI Agent platform. Our primary objective was to establish a robust MCP framework that allows AI Agents to dynamically interact with blockchain-based tools. From protocol design to server architecture, we optimized MCP’s core interaction model, setting the foundation for future expansion and cross-chain integration.

In this phase, we designed and implemented mechanisms for smart contract registration and ABI parsing. By pre-registering on-chain smart contracts and their ABI interfaces, the MCP server can dynamically recognize these on-chain functionalities and automatically generate corresponding endpoints. This means developers only need to provide the contract address and ABI details, and the platform seamlessly registers it as a usable ‘tool.’

Based on this foundation, we developed the MCP server code generator, which automatically creates service modules for interacting with these contracts. We also defined a minimal registration protocol, specifying how the MCP server registers its available tools to the AI Agent (Host). This lightweight protocol ensures a standardized exchange of capability lists and call requests between the Host and the Server.

As a result, AI Agents no longer need to preload all tools. Instead, they can dynamically discover and invoke external tools. Once a new smart contract service is registered, the Agent client learns of its capabilities through standard protocols and invokes the corresponding contract functions when required. This architecture significantly improves platform flexibility: tools can be added or removed without modifying the Agent itself — just register or deregister on the server side, and it’s ready for AI interaction. This solid infrastructure reinforced our confidence in MCP’s ability to drive closed-loop interactions between AI Agents and blockchain functionalities.

AI Agents (LLM in the Host client) communicate with the MCP server via the MCP protocol to call blockchain-based smart contract tools. The MCP server pre-registers these smart contracts and their ABI interfaces, enabling the Agent to dynamically discover and use these on-chain functionalities.

Phase 2: Expanding Web3 Access and SDK Integration

After establishing the core framework, the second phase focused on expanding Web3 access capabilities and providing developers with an SDK to support real-world application scenarios. In this phase, the capabilities of the MCP platform were significantly extended: the resources accessible by AI Agents were no longer limited to contract calls but broadened to encompass a wide array of Web3 data.

We implemented blockchain data query interfaces, enabling Agents to read on-chain state data such as account balances, transaction histories, and contract event logs. Previously, AI could only call contract functions to retrieve values; now, it can query blockchain data directly, allowing for smarter and more comprehensive decision-making. Additionally, we integrated decentralized storage systems, including IPFS and Arweave, enabling Agents to retrieve files based on Content IDs (CIDs) or store generated data on distributed networks. This improvement significantly expanded the Agent’s decentralized resource access capabilities, such as NFT metadata and social content.

Moreover, we introduced Oracle services to bridge off-chain real-world data — such as price feeds and weather information — into smart contract interactions. Through Oracle interfaces, AI Agents can reference dynamically updated off-chain data during contract execution, achieving seamless integration between real-world events and on-chain logic.

To support these new capabilities, we developed and released SDK v1.0, providing developers with a unified interface and toolset to leverage MCP’s features. The SDK abstracts low-level communication and data formatting complexities, allowing developers to interact with on-chain and off-chain resources seamlessly. SDK v1.0 supports common Web3 scenarios, including DeFi, NFTs, and decentralized identity, enabling developers to build AI Agent applications effortlessly.

Phase 3: Strengthening Security and Governance

As the platform entered deeper application stages, security and governance became critical focal points. In the early stages, as a less intelligent option, we had to tolerate a period of user authorization signatures. We have been closely monitoring the development of more secure AI execution environments. One promising area is the Trusted Execution Environment (TEE). TEE is an isolated execution environment that allows sensitive code to be executed securely without external interference. We believe that TEE could become a key part of MCP’s security strategy in the future, especially for handling private data and cross-chain assets.

However, TEE currently faces technical challenges, such as hardware dependencies and compatibility in multi-chain environments. Therefore, we are actively monitoring its progress, particularly in Distributed Trusted Computing. As these technologies mature, MCP’s security modules will be further enhanced, supporting higher security levels for smart contract calls and data access.

In addition, we established a highly secure multi-signature wallet management module, offering multi-layer authentication and permission control to ensure the safety of sensitive on-chain interactions. Based on the principle of least privilege, the platform enables developers to configure Agent invocation permissions, preventing unauthorized actions.

We also introduced a comprehensive identity authentication mechanism and auditing system, ensuring that every on-chain transaction is traceable and accessible for governance-level adjustments. In parallel, we deployed DAO governance mechanisms, enabling on-chain voting and decision-making through smart contracts, further enhancing platform transparency and security.

Phase 4: Ecosystem Growth and Multi-Chain Expansion

The final phase focuses on multi-chain support and ecosystem expansion. We introduced a standardized MCP tool registration and discovery mechanism, enabling AI Agents to invoke multi-chain operations and collaborate with diverse tools. The MCP network’s multi-chain support allows Agents to seamlessly interact with Ethereum, BNB Chain, Polygon, and Solana.

We launched multi-language SDKs (Python, TypeScript), and opened a decentralized tool registry. Developers can access multi-chain data, create custom tools, contribute to the open ecosystem, and earn on-chain rewards through an incentive mechanism. Additionally, we introduced ecosystem incentive programs, including developer grants, community governance rewards, and cross-chain task support, accelerating MCP’s global expansion.

As a core project of the BNB ecosystem, we believe that enriched cross-chain infrastructure is beneficial to BNB’s growth.

We have also started working on the onchain MCP proxy, allowing developers to register services with one click. Furthermore, we are exploring universal packaging methods that enable legacy contracts to be quickly understood and processed by AI Agents. As this progresses, REVOX Studio is set to become the largest MCP service layer.

We are also building towards the REVOX AI Agent Launchpad and Studio. REVOX Studio is the core AI Agent development platform in the BNB ecosystem. In the past, we used a plugin-based approach to achieve agent composition. We went through painful stages — you would never know how chaotic the interface fields defined by developers could be, and the chaos would double two months later. Fortunately, MCP gave us a solution. This solution requires developer consensus, and once it is achieved, it will unleash great power.

We are currently working on onchain MCP proxies to allow developers to complete service registration with one click. Furthermore, we are exploring universal packaging methods to enable legacy contracts to be quickly understood and processed by AI. As this development continues, REVOX Studio is set to become the largest MCP service layer.

Looking forward, Launchpad will be a unique component in the REVOX ecosystem tools. Launchpad signifies assetization and distribution, which is one of Web3’s fundamental innovations. We believe Launchpad is an essential component of the developer ecosystem, and REVOX will provide capitalization solutions for high-quality AI work. Meanwhile, Launchpad is also an indispensable operational tool and returns mechanism for community ecosystems. Therefore, Launchpad Service will be one of our main focuses — it will be a Service. Imagine what would happen if AI helped you participate in Launchpad. That would be interesting.