Nucleus MCP
Nucleus MCP provides a local, unified memory layer that synchronizes context and decisions across different AI tools like Cursor, Claude Desktop, and Windsurf. It enables cross-platform state management and persistent knowledge storage using a shared local brain.
README
🧠 Nucleus MCP
The Universal Brain for AI Agents — One brain that syncs Cursor, Claude Desktop, Windsurf, and any MCP-compatible tool.
🎯 The Problem
You use multiple AI tools daily:
- Cursor for coding
- Claude Desktop for thinking
- Windsurf for exploration
- ChatGPT for quick reasoning
But they don't share memory. (Until now).
Every time you switch tools, you lose context. You re-explain decisions. You repeat yourself constantly.
✨ The Solution
Nucleus syncs them with one brain.
Tell Claude about a decision → Cursor knows it
Make a plan in Windsurf → Claude remembers it
One brain. All your tools.
<!-- Demo video: https://github.com/eidetic-works/nucleus-mcp/releases - see demo_video.mp4 -->
🚀 What Makes Nucleus Different
| Feature | Other Solutions | Nucleus |
|---|---|---|
| Cross-Platform Sync | Single platform only | ✅ Syncs ALL your AI tools |
| Sovereignty | Cloud-dependent | ✅ 100% local, your data stays on your machine |
| Protocol | Proprietary | ✅ MCP standard (Anthropic-backed) |
| Security | Often misconfigured | ✅ Secure by default, audit logs included |
| Lock-in | Platform-specific | ✅ MIT license, open standard |
⚡ Quick Start (2 Minutes)
1. Install
pip install nucleus-mcp
2. Initialize
nucleus-init --scan
This creates your .brain/ folder, auto-configures Claude Desktop, and automatically ingests your README.md to seed the brain with initial context.
3. Restart Claude Desktop
Then try:
"What decisions have we made about the architecture?"
Claude will now remember across sessions!
🔧 Manual Configuration
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):
{
"mcpServers": {
"nucleus": {
"command": "python3",
"args": ["-m", "mcp_server_nucleus"],
"env": {
"NUCLEAR_BRAIN_PATH": "/path/to/your/project/.brain"
}
}
}
}
Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"nucleus": {
"command": "python3",
"args": ["-m", "mcp_server_nucleus"],
"env": {
"NUCLEAR_BRAIN_PATH": "/path/to/your/project/.brain"
}
}
}
}
Windsurf
Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"nucleus": {
"command": "python3",
"args": ["-m", "mcp_server_nucleus"],
"env": {
"NUCLEAR_BRAIN_PATH": "/path/to/your/project/.brain"
}
}
}
}
ChatGPT (Web)
- Go to Settings → Apps → Advanced → Developer Mode.
- Run the Nucleus SSE Bridge:
python scripts/sse_bridge.py. - Add
http://localhost:8000/sseas your MCP endpoint.
See the Community FAQ for more details.
🛠 Core Tools
Memory
| Tool | Description |
|---|---|
brain_write_engram |
Store persistent knowledge |
brain_query_engrams |
Retrieve knowledge |
brain_audit_log |
View decision history |
Sync (Multi-Agent)
| Tool | Description |
|---|---|
brain_sync_now |
Manually trigger brain sync |
brain_sync_status |
Check sync state and conflicts |
brain_sync_auto |
Enable/disable auto-sync |
brain_identify_agent |
Register agent identity |
State Management
| Tool | Description |
|---|---|
brain_get_state |
Get current project state |
brain_set_state |
Update project state |
brain_list_artifacts |
List all artifacts |
Hypervisor (Security)
| Tool | Description |
|---|---|
lock_resource |
Lock file/folder (immutable) |
unlock_resource |
Unlock resource |
watch_resource |
Monitor file changes |
hypervisor_status |
View security state |
🔄 Multi-Agent Sync (The Killer Feature)
Multiple agents, one brain.
# Agent A (Claude Desktop) makes a decision
brain_sync_now() # Syncs to shared .brain/
# Agent B (Cursor) automatically sees it
brain_sync_status() # Shows last sync, active agents
Features:
- Intent-Aware Locking — Files locked with WHO/WHEN/WHY metadata
- Conflict Detection — Last-write-wins with manual resolution option
- Auto-Sync — Optional file watcher for real-time sync
- Audit Trail — Every sync logged to
events.jsonl
⚔️ Comparison: Nucleus vs Alternatives
| OpenClaw | Claude Code | Nucleus | |
|---|---|---|---|
| What it syncs | OpenClaw → OpenClaw | Claude → Claude | Everything ↔ Everything |
| Cross-platform | ❌ | ❌ | ✅ |
| Local-first | ⚠️ Some cloud | ⚠️ Some cloud | ✅ 100% local |
| MCP Native | ❌ Custom protocol | ⚠️ Limited | ✅ Full MCP |
| Open Source | ✅ MIT | ❌ Closed | ✅ MIT |
| Security | ❌ Sleeper agents, key leaks | ⚠️ Cloud-managed | ✅ Hypervisor + audit trail |
| Cross-platform | ❌ | ❌ | ✅ |
| Local-first | ⚠️ Some cloud | ⚠️ Some cloud | ✅ 100% local |
OpenClaw trades security for capability. Nucleus gives you both. Nucleus connects ALL your platforms with one brain.
📁 The .brain/ Folder
Nucleus stores everything in a .brain/ folder in your project:
.brain/
├── config/
│ └── nucleus.yaml # Configuration
├── ledger/
│ ├── state.json # Current state
│ ├── events.jsonl # Audit log
│ └── decisions.md # Decision history
├── artifacts/
│ └── ... # Your stored knowledge
└── sessions/
└── ... # Saved sessions
Your data. Your machine. Your control.
🤝 Contributing
We're building the universal brain for AI agents. Join us!
- 🐛 Found a bug? Open an Issue
- 💡 Feature idea? Start a Discussion
- 🔧 Want to contribute? See CONTRIBUTING.md
✨ Contributors
A huge thank you to the pioneers helping build the universal brain:
- @aryasadawrate19 — Added Linux XDG support for
nucleus-init.
Development Setup
git clone https://github.com/eidetic-works/nucleus-mcp.git
cd nucleus-mcp
python3 -m venv venv
source venv/bin/activate
pip install -e ".[dev]"
pytest tests/
📜 License
MIT © Nucleus Team
⭐ Support
Star us on GitHub if Nucleus saves you from context amnesia!
One brain. All your AI tools. No more repeating yourself.
Built for the AI-native developer.
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。