mcp-evm

Project Introduction: EVM Chain Tools for AI Agents

This project implements a comprehensive EVM (Ethereum Virtual Machine) toolset designed specifically for AI agents to interact with blockchain networks through the ModelContext Protocol (MCP). Built with Bun runtime for superior performance, it leverages viem for Ethereum interactions and integrates secure key management, multi-network support, and standardized AI-agent communication interfaces.

Core Components

1. EVM Configuration Manager

- Manages RPC endpoints, chain IDs, and gas settings across networks (Mainnet/Sepolia supported)

- SQLite-backed storage with default network selection

- Extensible architecture for adding custom networks

2. Secure Wallet Management

- Encrypted private key storage with alias support

- Default wallet selection system

- Public address derivation and validation

3. Blockchain Interaction Tools

- Native ETH/ERC-20/ERC-721 balance queries

- Token transfer capabilities with gas estimation

- Transaction simulation using viem's public client

4. MCP Protocol Integration

- Standardized tool definitions using Zod schemas

- 9 pre-built tools for common EVM operations

- Error-handling framework for reliable AI agent interactions

5. HTTP Interface

- REST endpoints for key management

- JSON-based configuration updates

- CORS-ready secure routing

Technical Highlights

- Performance: Leverages Bun runtime + SQLite for sub-100ms cold-start operations

- Security: Private key isolation with separate encrypted database

- Extensibility: Modular design supporting quick integration of new EVM chains

- Interoperability: MCP-standardized API for cross-platform AI agent compatibility

- DevEx: Full TypeScript support with auto-generated JSON schemas

Current Implementation Status

- Functional ERC-20/ERC-721 transfers with simulation

- Working network configuration system

- Key management API endpoints

- MCP protocol compliance certification

- Test coverage for core transactions

Vision

We aim to become the default blockchain interface layer for AI agents, bridging natural language processing with on-chain operations. Future updates will introduce multi-chain atomic transactions, gas optimization AI models, and decentralized identity integrations. The project demonstrates how structured LLM tooling can enable trustworthy autonomous blockchain interactions while maintaining enterprise-grade security standards.

Next Phase Targets

1. Gas fee prediction engine using historical data

2. Cross-chain smart contract call support

3. Transaction monitoring webhooks

4. Hardware wallet integration

5. On-chain reputation system for AI agents

This implementation showcases a practical fusion of modern TypeScript tooling with blockchain fundamentals, creating a robust foundation for AI-driven Web3 applications.