SCS-MCP - Smart Code Search for Claude Desktop
An MCP (Model Context Protocol) server that provides intelligent semantic code search capabilities to Claude Desktop.
Features
- 🔍 Semantic Search - Find code by meaning, not just text matching
- 🧵 Thread-Safe - Handles concurrent requests without SQLite threading errors
- 🚀 Fast Performance - <100ms search response time with caching
- 📊 Multi-Language Support - Python, JavaScript, TypeScript, and more
- 🔄 Hybrid Search - Combines semantic and text matching for best results
Installation
Prerequisites
- Python 3.8+
- Claude Desktop
- WSL (for Windows users)
Quick Install
- Clone the repository:
git clone https://github.com/stevenjjobson/scs-mcp.git
cd scs-mcp
- Run the installer:
- Configure Claude Desktop by adding to
%APPDATA%\Claude\claude_desktop_config.json:
{
"mcpServers": {
"scs-mcp": {
"command": "wsl",
"args": [
"bash",
"-c",
"cd /path/to/scs-mcp && python3 -m src.server"
]
}
}
}
- Restart Claude Desktop
1. search
Search code by meaning using AI embeddings
Query: "database connection thread safety"
Returns: ThreadSafeDB class and related implementations
2. index
Index a project for semantic search
Project: /path/to/project
Force: true/false to reindex
3. analyze_symbol
Deep analysis of a specific symbol
Symbol: "SmartCodeSearch"
Returns: Definitions, references, and tests
4. find_similar
Find code similar to a given snippet
Code: "def test():"
Returns: Similar function definitions
5. get_context
Get relevant context for current work
Recent files: ["file1.py", "file2.py"]
Recent symbols: ["function1", "class1"]
Architecture
scs-mcp/
├── src/
│ ├── server.py # Main MCP server
│ ├── core/
│ │ ├── clean_search.py # Clean search implementation
│ │ ├── db_wrapper.py # Thread-safe SQLite wrapper
│ │ └── search.py # Search compatibility layer
│ └── utils/
├── scripts/
│ ├── install.sh # Installation script
│ └── download_model.py # Model downloader
├── config/
│ └── mcp_config.json # MCP configuration
└── requirements.txt
Key Improvements
- ✅ Fixed SQLite threading issues with thread-local connections
- ✅ Clean MCP protocol communication (no stdout pollution)
- ✅ Compatible with existing database schemas
- ✅ Handles both dictionary and tuple result formats
- ✅ Semantic search with proven accuracy
- First search: <100ms (model pre-loaded)
- Cached searches: <20ms
- Memory usage: ~500MB baseline
- Supports 100+ searches/minute
Development
To test the server locally:
python3 scripts/test_server.py
License
MIT
Author
Steven J. Jobson
Acknowledgments
Built for use with Claude Desktop using the Model Context Protocol (MCP).