code-context-mcp

An MCP (Model Context Protocol) server that provides semantic code search capabilities to AI assistants using MongoDB Atlas Vector Search and Voyage AI embeddings.

🚀 Features

• Semantic Code Search: Find relevant code based on meaning, not just keywords
• MongoDB Atlas Vector Search: Unified platform for vectors and data
• Voyage AI Embeddings: State-of-the-art code embeddings (MongoDB exclusive)
• Native Hybrid Search: Combine vector and text search with MongoDB's $rankFusion
• Real-time Sync: Automatic updates with MongoDB Change Streams
• Multi-language Support: TypeScript, JavaScript, Python, Java, Go, Rust, C++

📋 Prerequisites

1. MongoDB Atlas Account (Free tier available)

   • Sign up at: https://www.mongodb.com/cloud/atlas/register
   • Create a cluster and get your connection string

2. Voyage AI API Key (200M tokens free)

   • Sign up at: https://dash.voyageai.com/
   • Get your API key

🔧 Installation

For Claude Desktop

# Install globally
npm install -g code-context-mcp

# Or use npx (recommended)
npx code-context-mcp

Configuration

Add to your Claude Desktop configuration:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "code-context": {
      "command": "npx",
      "args": ["code-context-mcp"],
      "env": {
        "MONGODB_URI": "mongodb+srv://username:[e-mail]/",
        "VOYAGE_API_KEY": "va_xxx"
      }
    }
  }
}

For Other AI Assistants

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "code-context": {
      "command": "npx",
      "args": ["code-context-mcp"],
      "env": {
        "MONGODB_URI": "mongodb+srv://...",
        "VOYAGE_API_KEY": "va_xxx"
      }
    }
  }
}

Windsurf

Add to your Windsurf configuration:

{
  "mcpServers": {
    "code-context": {
      "command": "npx",
      "args": ["code-context-mcp"],
      "env": {
        "MONGODB_URI": "mongodb+srv://...",
        "VOYAGE_API_KEY": "va_xxx"
      }
    }
  }
}

🛠️ Available Tools

The MCP server provides the following tools to AI assistants:

index_codebase

Index a codebase for semantic search.

{
  "path": "/path/to/project",
  "name": "my-project"
}

search_code

Search for code semantically.

{
  "projectPath": "/path/to/project",
  "query": "authentication middleware",
  "limit": 10,
  "threshold": 0.7
}

get_file_content

Retrieve specific file content.

{
  "projectPath": "/path/to/project",
  "relativePath": "src/auth.ts",
  "startLine": 10,
  "endLine": 50
}

list_indexed_projects

List all indexed projects.

clear_index

Clear the index for a project.

{
  "projectPath": "/path/to/project"
}

get_project_stats

Get statistics about an indexed project.

{
  "projectPath": "/path/to/project"
}

⚙️ Configuration

Environment Variables

Variable	Description	Required	Default
MONGODB_URI	MongoDB Atlas connection string	✅	-
VOYAGE_API_KEY	Voyage AI API key	✅	-
MONGODB_DATABASE	Database name	❌	code_context
MONGODB_COLLECTION	Collection name	❌	embeddings
VOYAGE_MODEL	Voyage AI model	❌	voyage-3.5
BATCH_SIZE	Embedding batch size	❌	10
MAX_FILE_SIZE	Max file size (MB)	❌	10

Voyage AI Models

Model	Best For	Performance
voyage-context-3	RAG systems, long documents	+14.24% vs OpenAI
voyage-3-large	Highest accuracy	+9.74% vs OpenAI
voyage-3.5	General purpose (default)	+8.26% vs OpenAI
voyage-3.5-lite	High throughput	+6.34% vs OpenAI
voyage-code-3	Source code	Best for code

🏗️ MongoDB Atlas Setup

1. Create a Vector Search Index

In MongoDB Atlas:

1. Navigate to your cluster
2. Click "Search" → "Create Search Index"
3. Choose "Atlas Vector Search"
4. Use this configuration:

{
  "fields": [{
    "type": "vector",
    "path": "embedding",
    "numDimensions": 1024,
    "similarity": "cosine"
  }]
}

2. Enable Hybrid Search (Optional)

For MongoDB 8.0+, create both vector and text indexes:

{
  "mappings": {
    "fields": {
      "embedding": {
        "type": "knnVector",
        "dimensions": 1024,
        "similarity": "cosine"
      },
      "content": {
        "type": "string",
        "analyzer": "lucene.standard"
      }
    }
  }
}

📊 Performance

Based on 2025 benchmarks:

• Storage: 83% less than competitors with int8 quantization
• Accuracy: Up to 14.24% better retrieval than OpenAI
• Speed: Native hybrid search 30% faster
• Cost: 96% storage reduction with binary quantization

🔍 Example Usage in Claude

Once configured, you can use natural language commands:

"Index my TypeScript project at /Users/me/my-project"

"Search for authentication middleware in my-project"

"Show me the implementation of the UserService class"

"Find all database connection code"

"What files handle error logging?"

🐛 Troubleshooting

Connection Issues

1. MongoDB Connection Failed

   • Verify your connection string
   • Check IP whitelist in Atlas
   • Ensure cluster is running

2. Voyage AI Authentication Failed

   • Verify API key is correct
   • Check API key has credits

3. MCP Not Responding

   • Check logs in stderr
   • Verify environment variables
   • Restart AI assistant

Debug Mode

Set DEBUG=true in environment variables for verbose logging:

{
  "env": {
    "DEBUG": "true",
    "MONGODB_URI": "...",
    "VOYAGE_API_KEY": "..."
  }
}

🤝 Contributing

Contributions are welcome! Please see our Contributing Guide.

📄 License

MIT License - see LICENSE for details.

🔗 Links

• GitHub Repository
• MongoDB Atlas
• Voyage AI
• Model Context Protocol

---

Built with 💚 by MongoDB and Voyage AI