MongoDB + Voyage AI Docs MCP Server

🎯 MongoDB Acquired Voyage AI! This MCP server combines documentation from both MongoDB and Voyage AI, providing unified semantic search across both ecosystems.

First-of-its-kind multi-source documentation MCP (Model Context Protocol) server with semantic search powered by MongoDB Atlas Vector Search and Voyage AI's contextualized embeddings.

Features

• 🔍 Unified Search: Search across MongoDB AND Voyage AI documentation
• 🚀 MongoDB Atlas Vector Search: Leverages MongoDB's native vector capabilities
• 🧠 Voyage Context-3: Latest contextualized embeddings with 32K context window
• 🔗 Cross-Reference Intelligence: Understands relationships between MongoDB and Voyage docs
• 📚 Multi-Source Support: MongoDB docs, Voyage docs, Python SDK, TypeScript SDK
• 💻 Code Intelligence: SDK-aware code examples from both ecosystems
• 🔄 Incremental Updates: Efficient refresh from multiple GitHub repos
• 📊 Analytics: Track popular searches and improve results over time

Why This MCP Server?

Powered by Voyage Context-3 - The latest breakthrough in contextualized embeddings:

• 14.24% better than OpenAI-v3-large
• Cross-chunk reasoning - Understands relationships between document sections
• First 200M tokens FREE - Get started without cost
• Context-aware - Chunks from same document are encoded together

Unlike traditional keyword-based documentation search, this server understands:

• Intent: "How to make queries faster" finds optimization guides, indexing docs, and performance tips
• Context: Understands relationships between concepts
• Learning Paths: Suggests prerequisites and next steps
• Code Examples: Finds relevant code snippets with explanations

Prerequisites

• Node.js 18+
• MongoDB Atlas account (free tier works)
• Voyage AI API key (free tier available)
• Claude Desktop (for MCP integration)

Installation

1. Install globally via npm

npm install -g mongodocs-mcp

2. Set up environment variables

Create a .env file:

# MongoDB Atlas
MONGODB_URI=mongodb+srv://username:password@cluster.mongodb.net/
MONGODB_DB_NAME=mongodocs_mcp

# Voyage AI (voyage-context-3)
VOYAGE_API_KEY=your_voyage_ai_api_key_here
VOYAGE_MODEL=voyage-context-3  # Latest contextualized model
VOYAGE_DIMENSION=1024  # Options: 256, 512, 1024, 2048

3. Initialize the server

mongodocs-mcp init

This will:

• Test MongoDB and Voyage AI connections
• Create necessary directories
• Generate configuration files

4. Build the search index

mongodocs-mcp index

This process:

• Clones documentation from multiple sources:
  • MongoDB official docs
  • Voyage AI Python SDK
  • Voyage AI TypeScript SDK
  • Voyage AI web documentation
• Parses and chunks documents intelligently
• Generates contextualized embeddings using voyage-context-3
• Creates cross-references between MongoDB and Voyage docs
• Stores everything in MongoDB Atlas with vector indexes

Note: Initial indexing takes 45-90 minutes depending on your internet speed.

5. Configure Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "mongodb-docs": {
      "command": "mongodocs-mcp",
      "args": ["start"],
      "env": {
        "MONGODB_URI": "your_mongodb_uri",
        "VOYAGE_API_KEY": "your_voyage_api_key"
      }
    }
  }
}

Or generate the config automatically:

mongodocs-mcp config

Usage

Once configured in Claude Desktop, you can:

Search Documentation

MongoDB Queries:

• "How do I create compound indexes in MongoDB?"
• "Explain aggregation pipeline with examples"
• "Show me transaction best practices"

Voyage AI Queries:

• "How to use contextualized embeddings?"
• "Explain voyage-context-3 model"
• "Show me reranking examples"

Cross-Ecosystem Queries:

• "How to implement vector search with MongoDB and Voyage?"
• "Best practices for embeddings in Atlas Vector Search"
• "Compare embedding models for MongoDB"

Find Code Examples

• "Find Python examples for aggregation"
• "Show me JavaScript code for bulk operations"

Explore Concepts

• "What are the prerequisites for learning sharding?"
• "Find related topics to change streams"

CLI Commands

# Initialize the server
mongodocs-mcp init

# Build or update search index
mongodocs-mcp index [--full]

# Test search functionality
mongodocs-mcp search "your query"

# Show statistics
mongodocs-mcp stats

# Refresh documentation
mongodocs-mcp refresh

# Generate Claude config
mongodocs-mcp config

# Start server (for testing)
mongodocs-mcp start

Documentation Sources

Included Documentation

1. MongoDB Documentation (github.com/mongodb/docs)

   • Complete MongoDB manual
   • Tutorials and guides
   • API references

2. Voyage AI Python SDK (github.com/voyage-ai/voyageai-python)

   • Python client documentation
   • Code examples
   • API integration guides

3. Voyage AI TypeScript SDK (github.com/voyage-ai/typescript-sdk)

   • TypeScript/JavaScript client
   • React integration examples
   • Node.js guides

4. Voyage AI Web Docs (docs.voyageai.com)

   • Conceptual guides
   • Model documentation
   • Best practices

Architecture

Multi-Source Processing

┌─────────────────────────────────────┐
│     Documentation Sources           │
├─────────────────────────────────────┤
│  MongoDB Docs  │  Voyage Python SDK  │
│  Voyage TS SDK │  Voyage Web Docs    │
└──────┬──────────────────┬───────────┘
       │                  │
       ▼                  ▼
┌─────────────────────────────────────┐
│    Multi-Source Parser              │
│  • Extract cross-references         │
│  • Identify SDK examples            │
│  • Tag with source metadata        │
└──────────────┬──────────────────────┘
               │
               ▼
┌─────────────────────────────────────┐
│  Contextualized Embeddings          │
│  • voyage-context-3                 │
│  • Document-aware encoding          │
│  • Cross-source relationships       │
└──────────────┬──────────────────────┘
               │
               ▼
┌─────────────────────────────────────┐
│    MongoDB Atlas Vector Search      │
│  • Unified search index             │
│  • Source filtering                 │
│  • Cross-reference navigation       │
└─────────────────────────────────────┘

MongoDB Schema

{
  // Content
  content: "Document content...",
  embedding: [0.123, ...], // 2048-dim vector
  
  // Rich Metadata
  metadata: {
    title: "Compound Indexes",
    section: "Indexes",
    docType: "guide",
    version: "7.0",
    difficulty: "intermediate",
    tags: ["performance", "indexes"]
  },
  
  // Relationships
  relationships: {
    prerequisites: [...],
    related: [...],
    nextSteps: [...]
  },
  
  // Code Examples
  codeExamples: [{
    language: "javascript",
    code: "...",
    description: "..."
  }],
  
  // Search Optimization
  searchMetadata: {
    keywords: [...],
    concepts: [...],
    popularity: 0.85
  }
}

Vector Search Pipeline

1. Query Processing: Convert natural language to embedding
2. Hybrid Search: Combine vector similarity with text search
3. Contextual Ranking: Consider relationships and popularity
4. Result Enrichment: Add related topics and examples

Advanced Features

Semantic Search Modes

• Semantic: Pure vector similarity search using contextualized embeddings
• Hybrid: Combines vector and text search (default)
• Exact: Traditional text matching

Contextualized Embeddings

This server uses voyage-context-3 which encodes document chunks together, preserving:

• Inter-chunk context within documents
• Section relationships and flow
• Better understanding of references and dependencies

Filtering

{
  version: "7.0",
  docType: ["guide", "tutorial"],
  difficulty: "beginner",
  tags: ["aggregation"]
}

Relationship Traversal

The server understands document relationships:

• Prerequisites (what to read first)
• Related topics (similar concepts)
• Next steps (learning progression)

Performance

• Search latency: < 200ms (p95)
• Indexing speed: ~1000 docs/minute
• Memory usage: < 512MB
• Concurrent users: 100+

Development

Setup

# Clone repository
git clone https://github.com/yourusername/mongodocs-mcp
cd mongodocs-mcp

# Install dependencies
npm install

# Run in development
npm run dev

Testing

# Run tests
npm test

# Test search
npm run cli search "test query"

Building

# Build TypeScript
npm run build

# Package for distribution
npm pack

Troubleshooting

MongoDB Connection Issues

• Ensure your IP is whitelisted in Atlas
• Check connection string format
• Verify database user permissions

Voyage AI Issues

• Verify API key is valid
• Check rate limits (free tier: 50 req/min)
• Ensure model name is correct

Indexing Issues

• Ensure sufficient disk space for docs cache
• Check network connectivity to GitHub
• Verify MongoDB Atlas has vector search enabled

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Add tests for new features
4. Submit a pull request

License

MIT

Acknowledgments

• MongoDB for excellent documentation
• Voyage AI for powerful embeddings
• Anthropic for the MCP protocol

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions

Built with ❤️ for the MongoDB community