MongoDocs MCP

Semantic search for MongoDB and Voyage AI documentation using Voyage Context-3 embeddings and MongoDB Atlas Vector Search.

🚀 Features

• Voyage Context-3 Embeddings: State-of-the-art contextualized embeddings that understand document structure
• MongoDB Atlas Vector Search: Native vector search with $vectorSearch aggregation
• MongoDB + Voyage AI Docs: Search both MongoDB and Voyage AI documentation (since MongoDB acquired Voyage)
• Smart Query Expansion: Automatically expands MongoDB and Voyage AI terms for better recall
• Incremental Updates: Keep documentation current with efficient refresh mechanism
• Simple Setup: Interactive wizard guides you through configuration

📦 Installation

npm install -g mongodocs-mcp

🔧 Setup

Prerequisites

1. MongoDB Atlas Cluster (M10 or higher for vector search)
2. Voyage AI API Key (get it at voyageai.com)

Quick Start

1. Run the setup wizard:

npx mongodocs-mcp setup

2. Add to your MCP configuration (.mcp.json or Claude settings):

{
  "mcpServers": {
    "mongodocs": {
      "command": "npx",
      "args": ["mongodocs-mcp"]
    }
  }
}

🛠️ Available Tools

mongodb-semantic-search

Search MongoDB documentation using natural language:

Query: "how to create indexes"
Returns: Relevant documentation about indexing, performance optimization, index types

mongodb-find-similar

Find documentation similar to provided content:

Content: "db.collection.aggregate([...])"
Returns: Related aggregation documentation

mongodb-explain-concept

Get comprehensive explanations at different depth levels:

Concept: "sharding"
Depth: "intermediate"
Returns: Detailed explanation of MongoDB sharding

mongodb-refresh-docs

Update documentation database:

Mode: "incremental" | "full"

mongodb-status

Check system status and statistics.

🏗️ Architecture

MongoDB Docs (GitHub) + Voyage AI Docs (GitHub)
                    ↓
            Document Fetchers
                    ↓
        Smart Chunker (512 tokens, no overlap)
                    ↓
    Voyage Context-3 (contextualized embeddings)
                    ↓
        MongoDB Atlas Vector Search
                    ↓
                MCP Tools

🔑 Key Innovations

1. Contextualized Embeddings: Each chunk understands its place in the full document
2. Unified Search: Search across MongoDB and Voyage AI docs seamlessly (MongoDB acquired Voyage)
3. Query Expansion: Search for "index" also searches "indexes", "indexing", "createIndex"
4. Zero Overlap Chunking: More efficient than traditional overlapping chunks
5. Cloud-Native: Everything runs in MongoDB Atlas, no local storage needed

📊 Performance

• Search Latency: 50-200ms
• Embedding Quality: 1024-dimensional vectors
• Storage: ~500MB for complete MongoDB documentation
• Free Tier: First 200M Voyage tokens free

🔄 Updating Documentation

To refresh documentation:

# Incremental update (only changed files)
npx mongodocs-mcp refresh

# Full refresh (all documentation)
npx mongodocs-mcp refresh --full

🧪 Testing

Test your installation:

npx mongodocs-mcp test

📝 Environment Variables

Create a .env file with:

MONGODB_URI=mongodb+srv://...
VOYAGE_API_KEY=pa-...
MONGODB_DATABASE=mongodocs
INCLUDE_VOYAGE_DOCS=true  # Include Voyage AI documentation
GITHUB_TOKEN=ghp_...      # Optional: Avoid rate limits

🤝 Contributing

Contributions welcome! This project uses:

• TypeScript
• MongoDB Atlas Vector Search
• Voyage AI Context-3 embeddings
• Model Context Protocol (MCP)

📄 License

MIT

🙏 Acknowledgments

Built with:

• Voyage AI - Context-3 embeddings
• MongoDB Atlas - Vector Search
• Model Context Protocol - MCP framework

Made with ❤️ for the MongoDB and AI community