Self-learning AI standards system - learns from corrections and auto-updates CLAUDE.md
MCP Standards - Personal Memory for Claude
⚠️ ARCHIVED PROJECT: This repository is archived and provided for reference only. The project was experimental and served as a proof-of-concept for automatic preference learning with AgentDB vector memory. See ARCHIVE.md for complete details.
Make Claude remember YOUR preferences automatically. Zero config, zero manual steps.
🎯 What This Does
Stop repeating yourself to Claude. This MCP server learns your preferences automatically:
You: "Install pytest"
Claude: pip install pytest
You: "Actually, use uv not pip"
Claude: ✓ Remembered
Next session:
You: "Install requests"
Claude: uv pip install requests [automatic]
One correction. Forever remembered.
⚡ Quick Start (5 Minutes)
1. Install
git clone https://github.com/airmcp-com/mcp-standards.git
cd mcp-standards
# Install dependencies
npm install
# Setup AgentDB
npm run setup
2. Configure Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"mcp-standards": {
"command": "uv",
"args": [
"run",
"--directory",
"/ABSOLUTE/PATH/TO/mcp-standards",
"python",
"-m",
"mcp_standards.server_simple"
]
}
}
}
⚠️ Replace /ABSOLUTE/PATH/TO/ with your actual path!
3. Restart Claude Desktop
Quit and relaunch Claude Desktop.
4. Test It!
You: "Remember: use uv not pip"
Claude: ✓ Remembered: 'use uv not pip' (python)
You: "What do you remember?"
Claude: I remember you prefer:
- Use uv not pip for Python projects
That's it! You're done. 🎉
🧠 How It Works
Automatic Learning
Just correct Claude naturally - it learns automatically:
Session 1:
You: "Use uv not pip"
→ Auto-detected and stored in AgentDB
Session 2+:
You: "Install anything"
→ Claude uses uv automatically
What Gets Remembered
✅ Tool preferences: "use uv not pip", "prefer yarn over npm" ✅ Workflow patterns: "run tests before commit" ✅ Code style: "use TypeScript for new files" ✅ Project conventions: "follow PEP 8"
Categories (Auto-Detected)
python- Python/pip/uv preferencesjavascript- npm/yarn/pnpm preferencesgit- Git workflow preferencesdocker- Docker/container preferencestesting- Test framework preferencesgeneral- Everything else
📊 Features
| Feature | Status | |---------|--------| | Auto-detection | ✅ Detects "use X not Y" automatically | | Semantic search | ✅ <1ms with AgentDB (150x faster than SQLite) | | Cross-session | ✅ Preferences persist forever | | Zero config | ✅ Works out of the box | | 100% local | ✅ No cloud, all private | | Simple | ✅ 5-minute setup |
📖 Documentation
- Quick Start Guide - Detailed setup instructions
- Validation Checklist - Testing & troubleshooting
- Implementation Plan - Technical details
- Skills Guide - How to use in Claude
🛠️ MCP Tools Available
Personal Memory (Simple Version)
// Store preference
remember({
content: "use uv not pip",
category: "python"
})
// Search preferences
recall({
query: "package manager",
top_k: 5
})
// List all categories
list_categories()
// Get statistics
memory_stats()
Config Standards (Bonus)
// Generate minimal CLAUDE.md from project config files
generate_ai_standards({
project_path: ".",
formats: ["claude"]
})
🚀 Architecture
Simple & Fast
User corrects Claude
↓
Auto-detection hook triggers
↓
Stores in AgentDB (semantic vector memory)
↓
Next session: Claude queries AgentDB automatically
↓
Uses remembered preference
Technologies
- AgentDB - Ultra-fast vector memory (<1ms search)
- Python - MCP server (async)
- SQLite - Fallback storage
- MCP Protocol - Claude Desktop integration
- 100% Local - No cloud dependencies
📁 Project Structure
mcp-standards/
├── src/mcp_standards/
│ ├── agentdb_client.py # AgentDB wrapper
│ ├── hooks/auto_memory.py # Auto-detection
│ └── server_simple.py # Simple MCP server
├── tests/
│ └── test_simple_setup.py # Validation tests
├── docs/
│ ├── QUICKSTART_SIMPLE.md # Setup guide
│ ├── VALIDATION_CHECKLIST.md # Testing guide
│ └── SIMPLE_V2_PLAN.md # Technical details
├── scripts/
│ └── setup-agentdb.js # Setup script
├── .claude/skills/
│ └── remember-preferences.md # Claude skill
└── README.md # This file
Clean. Simple. Works.
🧪 Testing
Run automated validation:
python3 tests/test_simple_setup.py
Expected output:
✓ PASS: Directory Structure
✓ PASS: Required Files
✓ PASS: Module Imports
✓ PASS: AgentDB Client Init
✓ PASS: Auto Memory Patterns
Results: 5/5 tests passed
Status: Ready for dev testing 🚀
🐛 Troubleshooting
Setup fails
# Check Node.js version
node --version # Need v18+
# Install AgentDB manually
npm install -g agentdb
npx agentdb --version
Claude Desktop doesn't connect
# Check logs
tail -f ~/Library/Logs/Claude/mcp*.log
# Look for initialization messages
# Should see: "MCP Standards (Simple) initialized"
Preferences not remembered
Check that:
- Server is running (check Claude Desktop MCP status)
- Corrections use clear phrases ("use X not Y")
- AgentDB path exists:
~/.mcp-standards/agentdb
More help: See Validation Checklist
🎯 Performance
| Metric | Value | |--------|-------| | Setup time | <5 minutes | | Server startup | <2 seconds | | Search speed | <1ms (AgentDB HNSW) | | Detection | Real-time (async) | | Storage | <10ms | | Memory usage | ~50MB (embedding model) |
150x faster than SQLite. Zero lag.
🔒 Privacy
- ✅ 100% local - Everything stored in
~/.mcp-standards/ - ✅ No cloud - No external API calls
- ✅ No telemetry - No data collection
- ✅ Your data - You control everything
📝 What Changed (v2 Simple)
We removed all the complexity:
| v1 (Old) | v2 Simple (New) | |----------|-----------------| | Manual MCP calls (4-5 steps) | ✅ Automatic (zero steps) | | SQLite keyword search (50ms+) | ✅ AgentDB vector search (<1ms) | | No semantic matching | ✅ Semantic understanding | | Complex setup | ✅ 5-minute setup | | 6,000+ LOC | ✅ ~950 LOC |
Result: 80% less code, 100x better UX
🤝 Contributing
This is a personal side project. If you want to contribute:
- Try it yourself first
- Open an issue describing what you want to add
- Wait for feedback before writing code
Please don't: Submit large PRs without discussion first.
📜 License
MIT License - See LICENSE file
🙏 Credits
Built with inspiration from:
- AgentDB - Ultra-fast vector memory
- Context Engineering Guide - Minimal CLAUDE.md principles
- Gunnar's approach - Simple side projects that solve personal problems
💬 Questions?
Q: Why not just use v1? A: v1 requires 4-5 manual MCP calls per correction. v2 is zero-touch.
Q: Do I need AgentDB?
A: Yes, but it's installed automatically via npm run setup.
Q: Is my data private?
A: 100% local. Everything stored in ~/.mcp-standards/. No cloud.
Q: What if I want the old version?
A: Use src/mcp_standards/server.py instead of server_simple.py.
🚀 Next Steps
# Try it now
git clone https://github.com/airmcp-com/mcp-standards.git
cd mcp-standards
npm run setup
# See: docs/QUICKSTART_SIMPLE.md
Stop repeating yourself. Start remembering automatically. 🎯
📦 Archive Status
This project is archived. See ARCHIVE.md for:
- Complete project status and achievements
- Security audit results
- Full documentation index
- Lessons learned and technical insights
- How to use this repository as reference
Made with ❤️ by keeping it simple