Memory Bank MCP Server

A MongoDB-powered Memory Bank MCP (Model Context Protocol) server that creates an intelligent, project-aware memory system for AI assistants. Built with MongoDB Atlas Vector Search's hybrid search capabilities.

What It Does

This MCP server provides AI assistants with persistent memory across sessions by:

• Storing structured project documentation in MongoDB
• Using vector embeddings for semantic search
• Combining vector and text search for powerful hybrid queries
• Maintaining strict project isolation
• Providing tools for reading, updating, and searching memories

Features

• 🧠 Intelligent Memory System - Structured documentation that evolves with your project
• 🔍 Hybrid Search - Combines MongoDB's vector and text search capabilities
• 🏗️ Project Isolation - Each project maintains its own memory space
• 🚀 High-Performance - Leverages MongoDB indexes and Voyage AI embeddings
• 🔧 MCP Integration - Works seamlessly with Claude and other MCP-compatible AI assistants

Installation

Prerequisites

• Node.js 18+
• MongoDB Atlas cluster with Vector Search enabled
• Voyage AI API key

Quick Start (Recommended)

Run directly with npx (no installation needed):

# Set environment variables
export MONGODB_URI="your-mongodb-uri"
export VOYAGE_API_KEY="your-voyage-key"

# Run the server
npx mongodb-memory-bank-mcp

Or configure in Claude's MCP settings:

{
  "mcpServers": {
    "memory-bank": {
      "command": "npx",
      "args": ["mongodb-memory-bank-mcp"],
      "env": {
        "MONGODB_URI": "your-mongodb-uri",
        "VOYAGE_API_KEY": "your-voyage-key"
      }
    }
  }
}

Local Development

1. Clone the repository:

git clone https://github.com/romiluz13/mongodb-memory-bank-mcp.git
cd mongodb-memory-bank-mcp

2. Install dependencies:

pnpm install

3. Configure environment:

cp .env.example .env.local
# Edit .env.local with your MongoDB URI and Voyage AI key

4. Build the project:

pnpm build

5. Create MongoDB indexes:

pnpm db:indexes

Running Locally

Development Mode

pnpm dev

Testing with MCP Inspector

pnpm mcp:inspect

Running Tests

pnpm test

Usage with Claude

Add to your Claude MCP settings:

{
  "mcpServers": {
    "mongodb-memory-bank-mcp": {
      "command": "npx",
      "args": ["mongodb-memory-bank-mcp-v2"],
      "env": {
        "MONGODB_URI": "your-mongodb-uri",
        "VOYAGE_API_KEY": "your-voyage-key"
      }
    }
  }
}

Version 1.3.0 - Context Engineering REVOLUTION! 🚀

Feature Blueprints - Your New SUPERPOWER!

• 🎯 Validation Gates: Ensure production-ready code EVERY time
• 📊 Confidence Scoring: Know exactly when to ship (0-10 scale)
• 🔍 Pattern Discovery: Find similar features with MongoDB hybrid search
• ✅ Multi-Level Validation: TypeScript → Tests → Integration → Production
• 🧠 Context Engineering: 100x better than prompt engineering!

Quick Start with Feature Blueprints

# Create a new feature blueprint
memory_bank/feature --action create --name "user-authentication"

# Validate your implementation
memory_bank/feature --action validate --name "user-authentication"

# Find similar patterns
memory_bank/feature --action search --name "auth"

Version 1.2.0 - AI-Optimized Memory Bank!

v1.2.0 Features - The Final 5%!

• 🎯 Smart Search Previews: Shows actual matching context, not just first 200 chars
• 🧠 AI-Focused Templates: Rich prompts helping AI understand project context better
• 🔗 Auto-Reference Detection: Automatically finds relationships between memory files
• 📊 Content Stats: Shows file size and version in search results
• ⏰ Access Tracking: Tracks last accessed time for freshness
• 💡 Better Memory Structure: Each file now has AI-specific guidance sections

v1.1.1 Fixes

• Fixed Hybrid Search Scoring: Proper handling of $rankFusion RRF scores
• Score Safety: All score values now safely handle undefined cases
• 100% Search Coverage: Vector, text, and hybrid all working perfectly

v1.1.0 Features

• MongoDB $rankFusion: Official MongoDB 8.1 operator for hybrid search
• Reciprocal Rank Fusion: Intelligent result ranking algorithm
• Weighted Scoring: Configurable weights (default: 70% vector, 30% text)
• Atlas Search Integration: Full-text search with Atlas Search indexes

v1.0.1 Fixes

• Fixed vector search index naming mismatch
• Fixed hybrid search filter configuration
• Improved search reliability

Memory Structure

The memory bank uses six core files:

• projectbrief.md - Foundation document
• productContext.md - Why the project exists
• activeContext.md - Current work focus
• systemPatterns.md - Architecture decisions
• techContext.md - Technology stack
• progress.md - What works and what's next

Contributing

See CONTRIBUTING.md for development guidelines.

Documentation

• CLAUDE.md - Detailed project context and rules
• WORKFLOWS/ - Reusable automation recipes
• ONBOARDING.md - Getting started guide

License

MIT