MongoDB Memory Bank MCP Server

A Model Context Protocol (MCP) server that provides persistent memory storage using MongoDB as the backend. This allows AI assistants to store, retrieve, and search through project-specific documentation and notes across sessions.

What This Does

This MCP server enables AI assistants to:

• Store and retrieve project documentation and notes
• Automatically organize memories by project with complete isolation
• Search through stored content using text search or semantic search (Atlas only)
• Maintain structured documentation templates (like projectbrief.md, activecontext.md)
• Auto-generate 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.

1. Create a MongoDB Atlas cluster (free tier available)
2. Get your connection string from Atlas
3. Sign up for Voyage AI API key for embeddings
4. Configure environment variables (see Configuration section)

Option 2: Local MongoDB Community

Provides core functionality with text search.

1. Install and start MongoDB Community Edition locally
2. Use connection string: mongodb://localhost:27017
3. 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 bank
• list_project_files - List all files within a specific project
• memory_bank_read - Read the content of a specific memory file
• memory_bank_write - Create or update a memory file
• memory_search - Search across all memories with text or semantic search

Advanced Features

• memory_discover - Find memories related to a specific file
• detect_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

1. Create projectbrief.md - System detects this as a project brief template
2. Create activecontext.md - System automatically creates missing systempatterns.md and techcontext.md if they don't exist
3. Update activecontext.md - System can automatically update progress.md based on changes
4. 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

1. Fork the repository
2. Create a feature branch
3. Add tests for new features
4. Ensure all tests pass
5. 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