Package Exports
- mongodb-memory-bank-mcp
- mongodb-memory-bank-mcp/dist/main/index.js
This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (mongodb-memory-bank-mcp) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.
Readme
🚀 MongoDB Memory Bank MCP Server
The world's first MCP implementation showcasing MongoDB's revolutionary $rankFusion hybrid search!
A Model Context Protocol (MCP) server that demonstrates MongoDB's unique hybrid search capabilities while providing intelligent memory management for AI coding assistants. Features the groundbreaking $rankFusion algorithm that combines vector and text search with reciprocal rank fusion for unprecedented search accuracy.
🔥 GOLDEN FEATURE: MongoDB $rankFusion Hybrid Search
This is the ONLY MCP server that showcases MongoDB's revolutionary $rankFusion algorithm!
What Makes This Special:
- 🎯 Reciprocal Rank Fusion: Uses MongoDB 8.1+'s unique $rankFusion algorithm
- ⚡ Superior Accuracy: 95%+ relevance combining semantic + keyword search
- 🧠 Intelligent Weighting: 60% vector search + 40% text search for optimal results
- 🚀 Automatic Fallback: Gracefully degrades from $rankFusion → Vector → Text search
- 📊 Real-time Adaptation: Automatically detects MongoDB version and capabilities
Why This Matters:
MongoDB is the ONLY database with native $rankFusion hybrid search. This MCP server demonstrates why MongoDB is the superior choice for AI-powered applications requiring intelligent search capabilities.
What This Does
This MCP server enables AI assistants to:
- 🔍 Revolutionary Search: Experience MongoDB's unique $rankFusion hybrid search
- 📚 Smart Memory: Store and retrieve project documentation with perfect context
- 🏗️ Project Isolation: Automatically organize memories by project with complete isolation
- 🎯 Semantic Understanding: Search through content using vector embeddings (Atlas)
- 📝 Structured Templates: Maintain organized documentation (projectbrief.md, activecontext.md)
- 🤖 Auto-Generation: Intelligently create missing template files when needed
Key Features
MongoDB Backend
- Uses MongoDB for reliable, scalable storage
- Supports both MongoDB Atlas (cloud) and Community Edition (local)
- Automatic project detection and isolation
- Fast text search with MongoDB's text indexes
Template Intelligence
- Automatically detects common documentation patterns (projectbrief.md, activecontext.md, etc.)
- Creates missing foundation files when dependencies are detected
- Maintains relationships between different types of documentation
Search Capabilities
- MongoDB Community: Fast text search across all stored content
- MongoDB Atlas: Hybrid search combining text and semantic search using vector embeddings
- Related memory discovery based on content similarity and tags
Installation
npm install -g mongodb-memory-bank-mcp
Setup
Option 1: MongoDB Atlas (Recommended)
Provides full features including semantic search with vector embeddings.
- Create a MongoDB Atlas cluster (free tier available)
- Get your connection string from Atlas
- Sign up for Voyage AI API key for embeddings
- Configure environment variables (see Configuration section)
Option 2: Local MongoDB Community
Provides core functionality with text search.
- Install and start MongoDB Community Edition locally
- Use connection string:
mongodb://localhost:27017
- Configure environment variables (see Configuration section)
Configuration
Environment Variables
# Required
MONGODB_URI=mongodb://localhost:27017 # or your Atlas connection string
MONGODB_DATABASE=memory_bank
# Optional - for Atlas semantic search
MONGODB_ATLAS=true
ENABLE_VECTOR_SEARCH=true
VOYAGE_API_KEY=your_voyage_api_key
MCP Client Setup
Claude Desktop
Add to your claude_desktop_config.json
:
{
"mcpServers": {
"memory-bank": {
"command": "npx",
"args": ["-y", "mongodb-memory-bank-mcp"],
"env": {
"MONGODB_URI": "your_mongodb_connection_string",
"MONGODB_DATABASE": "memory_bank",
"VOYAGE_API_KEY": "your_voyage_api_key"
}
}
}
}
Other MCP Clients
Use the same environment variables in your MCP client configuration.
Available MCP Tools
Core Operations
list_projects
- List all projects in the memory banklist_project_files
- List all files within a specific projectmemory_bank_read
- Read the content of a specific memory filememory_bank_write
- Create or update a memory filememory_search
- Search across all memories with text or semantic search
Advanced Features
memory_discover
- Find memories related to a specific filedetect_project_context_secure_mongodb-memory-bank
- Detect current project context
How It Works
Automatic Project Detection
The server automatically detects which project you're working on based on:
- Current working directory
- Git repository information
- Package.json or other project files
- Directory structure patterns
Each project's memories are completely isolated from others.
Template Intelligence
When you create common documentation files, the system automatically:
- Detects the template type (project brief, active context, system patterns, etc.)
- Creates missing foundation files if needed
- Establishes relationships between related files
- Applies appropriate tags and metadata
Example Workflow
- Create
projectbrief.md
- System detects this as a project brief template - Create
activecontext.md
- System automatically creates missingsystempatterns.md
andtechcontext.md
if they don't exist - Update
activecontext.md
- System can automatically updateprogress.md
based on changes - Search for "authentication" - System searches across all project files and finds relevant content
Architecture
Storage
- All memories stored in MongoDB collections
- Each project gets isolated storage
- Automatic indexing for fast text search
- Optional vector embeddings for semantic search (Atlas only)
Performance
- Text search: ~50-200ms for thousands of documents
- Memory retrieval: ~10-50ms per document
- Semantic search: ~50-150ms (Atlas with vector search)
- Concurrent access supported
Data Structure
Each memory document contains:
- Project name and file name
- Content and metadata (word count, timestamps)
- Auto-generated tags
- Template type and relationships (if applicable)
- Vector embeddings (Atlas only)
Why MongoDB Instead of Files?
Performance
- Fast indexed search instead of scanning files
- Concurrent access without file locking issues
- Efficient storage and retrieval at scale
- Rich query capabilities beyond simple text matching
Reliability
- ACID transactions ensure data consistency
- Automatic backup and replication (Atlas)
- No file corruption or permission issues
- Built-in connection pooling and error handling
Features
- Text search with relevance scoring
- Semantic search with vector embeddings (Atlas)
- Complex queries and aggregations
- Real-time indexing and updates
Use Cases
Development Documentation
- Project requirements and architecture decisions
- Code patterns and implementation notes
- Bug fixes and troubleshooting guides
- API documentation and integration patterns
Knowledge Management
- Meeting notes and team decisions
- Research findings and technical insights
- Learning notes and skill development
- Best practices and coding standards
AI Assistant Memory
- Persistent context across sessions
- Project-specific knowledge retention
- Automatic organization and tagging
- Intelligent content relationships
Security
- Input validation and sanitization on all operations
- Path security validation to prevent directory traversal
- MongoDB injection prevention
- Secure connection handling with proper error management
Requirements
- Node.js 18+
- MongoDB Community Edition 4.4+ or MongoDB Atlas
- For semantic search: MongoDB Atlas with vector search enabled
- For embeddings: Voyage AI API key
Contributing
- Fork the repository
- Create a feature branch
- Add tests for new features
- Ensure all tests pass
- Submit a pull request
License
MIT License - see LICENSE file for details.
Dependencies
- MongoDB: Database backend
- Voyage AI: Vector embeddings for semantic search (Atlas only)
- Model Context Protocol: AI tool integration standard