Ethereum Code

AI-native CLI that takes founders from raw idea to deployed Ethereum dApp in hours, not weeks.

Ethereum Code

Created At

Open Agents

Project Description

ethereum-code is a single TypeScript CLI called eth that orchestrates Claude-powered AI agents through every phase a founder hits on the way from napkin sketch to deployed, audited, and funded product on Ethereum or its L2s. Run eth new and answer one question. The tool handles ideation, architecture, smart contract generation, security auditing, deployment, and fundraising prep automatically. The core innovation is grounding. Every agent loads 31 skill documents bundled locally before writing a single line of code. No hallucinated contract addresses. No stale gas assumptions. No generic advice when your use case clearly belongs on Base or Arbitrum. Alongside this, the tool ships with a curated catalog of 88 cloneable Ethereum repos across 8 categories so builders never start from zero, 29 MCP servers for extending the workflow with external integrations, and a corpus of 52 fundable Ethereum ideas growing to 500. The five engines cover the full founder journey from raw idea to investor deck, all chain aware and grounded in real documentation.

How it's Made

The CLI is built in TypeScript running on Node 20, bundled with tsup into a single ESM executable. The agent runtime is the single chokepoint for every Claude call. It looks up the task, loads the relevant skill markdown from 31 bundled SKILL.md files covering everything from L2 selection and gas optimization to frontend UX and audit checklists, injects them as system context, and enforces hard rules: no hallucinated addresses, live gas checks via cast base-fee, and no secrets in any file writes. The 88 curated repos are searchable via eth repos and browseable across categories like DeFi, NFTs, DAOs, ZK, and account abstraction. The 29 MCP servers listed in eth mcps let teams plug in external tools like Asana, GitHub, or on-chain data sources directly into their agent workflows without manual wiring. Skill routing is handled by a registry table that maps each task to the exact subset of the 31 skills needed, keeping context lean and agent outputs precise. Claude Opus 4.6 powers the architect, auditor, and reviewer agents. Claude Sonnet handles the builder and iteration loops. The auditor is always a separate agent run from the builder so it never reviews its own code. The reference template is a production grade ERC-4626 vault with fuzz tests, a 48 hour timelocked strategy rotation, and a three button Scaffold-ETH 2 frontend using exact allowances only.

background image mobile

Join the mailing list

Get the latest news and updates