srclight

Srclight is a Python-based MCP server that builds a rich, local code index and exposes a suite of MCP-compatible tools for AI coding agents. It provides incremental code indexing, symbol and relationship querying, semantic search via embeddings, and multi-repo workspace support, all accessible through the MCP interface. The server offers 29 tools for symbol search, graph relationships, change intelligence, and document extraction, enabling agents to operate with a detailed and searchable understanding of your codebase. You can run the server in standard stdio mode or via SSE for persistent sessions, and you can integrate it with Claude Code, Cursor, or any MCP client.

Typical workflow: index your project with srclight index to build the local SQLite databases, optionally embed symbols with embeddings using Ollama or other providers, and then serve the MCP endpoint with srclight serve so clients can connect and issue mcp add, mcp query, and other MCP actions. Srclight also supports multi-repo workspaces, so you can attach and search across multiple repositories in a single session. The server exposes commands for indexing, searching, symbol lookup, and workspace management, plus extended capabilities like semantic search and document extraction for broader code understanding.

Prerequisites:

• Python 3.11+
• Git
• Optional: Ollama for embeddings, CUDA-capable GPU for GPU-accelerated vector search

Install from PyPI (recommended):

1. Create a virtual environment (optional but recommended): python3 -m venv venv source venv/bin/activate
2. Install Srclight from PyPI: pip install srclight

Install from source (optional):

1. git clone https://github.com/srclight/srclight.git
2. cd srclight
3. pip install -e .

Optional extras (for full feature set):

• Embeddings / embeddings-based semantic search: pip install 'srclight[all]'
• PDF, DOCX, images support and OCR: pip install 'srclight[docs,pdf]'
• GPU acceleration (CUDA required): pip install 'srclight[gpu]'

Usage basics:

• Index a project: srclight index
• Start MCP server: srclight serve
• Embedding setup (optional): ensure Ollama is running and index with embeddings, e.g. srclight index --embed qwen3-embedding

Tips:

• The index and embedding data live under .srclight/ in your project root. Do not commit these to version control.
• For multi-repo workspaces, initialize a workspace and add repos, then index and serve within the workspace context.

Notes and tips:

• Srclight requires Python 3.11+ and Git. Optional components include Ollama for embeddings, PaddleOCR for scanned PDFs, and GPU acceleration via CUDA for faster vector search.
• You can run the MCP server in stdio mode or via SSE. SSE is recommended for multiple sessions and workspaces.
• Embeddings are stored in symbol_embeddings within the index database. Ensure sufficient disk and, if using GPU, available VRAM.
• If you use submodules, they are not indexed automatically; index submodules by adding them as separate workspaces.
• For best performance with large codebases, enable incremental reindexing and consider using multi-repo workspaces to consolidate search across repos.