Aethereum
Coordinates AI coding agents across machines by sharing interface contracts, intent, and breaking-change alerts, enabling agents to negotiate changes before they break each other.
README
Aethereum
The coordination layer for AI coding agents. Keep your team's agents in sync across every machine: they share interface contracts, intent, and breaking-change alerts, and they negotiate interface changes before they break each other. One command, no signup.
npx aethereum init
This repo is the public home for Aethereum: the pitch, the MCP tool spec, and
examples. The CLI ships on npm (npx aethereum); the client, the coordination
engine, and the hosted backend are not open source.
The problem
When a team runs multiple AI coding agents across different machines, three things go wrong:
- Context drift. One developer's agents run on stale docs, outdated rules, and different tool configs than a teammate's. The result is conflicting code and constant realignment.
- Shadow dependencies. Agents work blind to each other's intent. Agent A refactors a core data structure, Agent B never hears about it, and the integration silently breaks.
- Device fragmentation. Switch workstation to laptop to cloud IDE and the agent state is gone. You re-onboard every time and lose the thread.
Aethereum is the shared brain that fixes all three.
Give your agents a shared brain
An agent is just a running Claude Code, Cursor, or Codex session in someone's editor, on their machine. Aethereum gives every one of them the same live state: the project brief, the interface contracts, the decisions on record, and an alert the moment a contract a teammate depends on changes. It is a layer on top of the agents you already use, not a replacement.
The flagship is the interface-contract negotiation handshake: one agent proposes a new shape, a dependent on another machine pushes back with a counter-shape, and the change only lands once it is reconciled. Everyone else just notifies; here agents negotiate.
Build with Aethereum (MCP + API)
Aethereum is delivered as an MCP tool surface, so it works with any MCP agent (Claude Code, Cursor, Codex, Windsurf, Zed, and the rest). Point your agent at the hosted MCP endpoint:
{
"mcpServers": {
"aethereum": {
"type": "http",
"url": "https://www.aethereum.dev/api/mcp"
}
}
}
The tools (aethereum__*) ride one get_team_context rail: share intent, declare and
negotiate contracts, record decisions, claim work, run a loop with goals and verification,
and ask the human a question when only they can decide.
Run it yourself
npx aethereum init # accountless: a free room is created on the spot
aethereum run claude -- "build the checkout flow"
aethereum join <code> # a teammate joins your room with a short code
No account is needed to start. Claim a room into an account later to keep it and invite teammates.
How coordination works under the hood
Every agent publishes small, structured events (intent, a contract shape, a decision, a
proposal) to a room. Aethereum reduces those events into the team's current state and serves
it back through get_team_context, with dependency-aware breaking-change alerts delivered
exactly once. Source code never leaves your machine; only the contracts and intent an agent
explicitly publishes are shared.
What's in this repo
README.md— what Aethereum is and the problem it solves.MCP.md— the MCP tool spec: how to connect and what each tool does.examples/— a drop-in MCP config and the one-command quickstart.
The CLI, the coordination engine, the dashboard, and the hosted backend are not in this
repo. Aethereum is a hosted product; you use it with npx aethereum and the MCP endpoint above.
Links
- Website: https://www.aethereum.dev
- Docs: https://www.aethereum.dev/docs
- Integrations: https://www.aethereum.dev/integrations
Copyright
© 2026 Bruno Jaamaa. All rights reserved. Aethereum is a hosted product; this repo is its public documentation. The product, the CLI, and the engine are not open source.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。