🧠 Memory MCP - Living Brain for AI Coding Assistants

Revolutionary MCP server that creates a "living brain" for each project - one document that evolves, learns, and gets smarter with every interaction.

✨ Features

• Perfect Project Isolation - SHA-256 hashed project paths, zero data leakage
• Universal AI Assistant Support - Works with Cursor, Augment, Claude, Windsurf
• MCP Protocol 2025-06-18 Compliant - Latest security and features
• MongoDB Fallback Strategies - Works on ANY MongoDB version (8.1+ to 4.4+)
• Sub-100ms Response Times - Ultra-optimized indexes and caching
• Enterprise Security - User consent, rate limiting, input validation

🚀 Quick Start

Installation

npm install -g memory-mcp

Prerequisites

• Node.js 18+
• MongoDB (local or Atlas)
• AI coding assistant (Cursor, Claude, etc.)

Configuration

Create a .env file:

MONGODB_URI=mongodb://localhost:27017/memory_mcp
# or for Atlas:
# MONGODB_URI=mongodb+srv://username:password@cluster.mongodb.net/memory_mcp

Run the Server

memory-mcp

🔧 Technical Resources

Schemas (Complete Zod Definitions)

• schemas/brain.schema.ts - Brain document structure
• schemas/mcp.schema.ts - MCP protocol interfaces

Examples (Working Code)

• examples/brain-operations.ts - Core brain operations
• examples/mcp-tool.ts - MCP server implementation
• examples/init-brain.ts - Brain initialization

Templates

• templates/initial-brain.json - Starting brain structure

🚀 Key Discoveries

The Breakthrough: Natural Language Interface

• Changed from 5 technical tools to 1 natural "brain" tool
• Usage jumped from 33% to 88%
• AI assistants now naturally say "let me check with your brain"

The Architecture: Living Brain Document

• ONE 3-5MB document per project (not thousands of small ones)
• Self-organizing knowledge graph
• Automatic consolidation and learning
• Uses MongoDB $rankFusion for intelligent search

The Psychology: How AIs Think

• Technical descriptions are ignored
• Natural language is embraced
• Contextual hints guide behavior
• Positive feedback increases usage

💡 Implementation Overview

// Instead of this (Memory Engineering v2):
await memory.update({ 
  type: 'working', 
  content: { action: 'fixed bug' } 
});

// We have this (Memory MCP):
brain "I fixed the login timeout with refresh tokens"
// AI responds: "✅ Learned! I'm 15% smarter about auth now!"

📊 Results

🎯 Mission Success Criteria

When implemented correctly:

1. AI naturally uses "brain" in conversation
2. Feels like collaboration with intelligent partner
3. Knowledge accumulates and connects
4. Patterns emerge automatically
5. Development accelerates over time

🛠️ For Claude Code

Dear Claude Code,

This planning represents hundreds of hours of learning. The key insight: make it feel natural, not technical. When implementing:

1. Start with 00_START_HERE.md
2. Keep the interface simple (one "brain" tool)
3. Use natural language everywhere
4. Test with real AI usage, not curl commands
5. Make responses conversational and encouraging

The goal isn't to build a database - it's to create a living, learning partner that makes you smarter with every interaction.

📚 Research References

Our journey included:

• MCP Protocol deep dive
• MongoDB Atlas Search mastery
• AI psychology research
• Community feedback analysis
• Extensive real-world testing

All distilled into these planning documents.

"We're not storing memories. We're growing intelligence."

🤝 Credits

Built on insights from:

• Memory Engineering v2.0 (the foundation)
• MongoDB team (for $rankFusion)
• MCP community (for protocol clarity)
• Anthropic (for Claude and MCP)
• Countless hours of testing and iteration

Ready to build something magical? Start with 00_START_HERE.md!