srclight and code-memory
These two tools are competitors, as both are offline, self-hosted MCP servers designed for deep codebase indexing and semantic search to empower AI agents, offering similar functionalities like vector search, Git history analysis, and code graph generation.
About srclight
srclight/srclight
Deep code indexing MCP server for AI agents. 25 tools: hybrid FTS5 + embedding search, call graphs, git blame/hotspots, build system analysis. Multi-repo workspaces, GPU-accelerated semantic search, 10 languages via tree-sitter. Fully local, zero cloud dependencies.
Implements a filesystem-watched SQLite + tree-sitter indexing layer that increments on git changes via post-commit hooks, with optional Ollama embeddings cached as `.npy` matrices for GPU-accelerated cosine similarity in ~3ms. Operates as a stdio or SSE MCP transport server compatible with Claude Code and Cursor, supporting multi-repo workspaces through SQLite ATTACH and RRF-ranked hybrid search across keyword (FTS5/trigram) and semantic results.
About code-memory
kapillamba4/code-memory
MCP server with local vector search for your codebase. Smart indexing, semantic search, Git history — all offline.
Implements a three-tier retrieval strategy (BM25 + dense vectors via SQLite-vec + AST parsing) across specialized tools that route queries by type—code definitions, architectural documentation, and Git history—powered by local sentence-transformers embeddings. Integrates as an MCP server with Claude Desktop, VS Code (Copilot/Continue), Gemini CLI, and Claude Code, supporting 8 languages with full AST extraction and 9+ with fallback whole-file indexing.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work