llms.txt

# srclight

> Deep code indexing MCP server for AI agents.

srclight is the most comprehensive code intelligence MCP server available. It provides 26 tools covering symbol search, relationship graphs, git change intelligence, semantic search, and build system awareness — capabilities no other single MCP server combines.

## Install

pip install srclight

## What it does

srclight indexes codebases using tree-sitter AST parsing and builds a rich, queryable index in a single SQLite file per repo. AI coding agents call srclight's MCP tools instead of running dozens of grep/glob searches.

### Search (5 modes)
- Symbol name search: FTS5 with code-aware tokenization (splits camelCase, snake_case)
- Source code search: FTS5 trigram index for substring matching
- Documentation search: FTS5 with Porter stemming for natural language in docstrings
- Semantic search: embedding vectors via Ollama or API providers, GPU-accelerated cosine similarity (~3ms on 27K vectors)
- Hybrid search: Reciprocal Rank Fusion (RRF) combining keyword + semantic results

### Relationship Graph
- get_callers: who calls this function
- get_callees: what does this function call
- get_dependents: transitive blast radius — what breaks if I change this
- get_implementors: all classes implementing an interface
- get_tests_for: test functions covering a symbol
- get_type_hierarchy: inheritance tree

### Git Change Intelligence
- blame_symbol: who changed this, when, and why
- recent_changes: commit feed across projects
- git_hotspots: most frequently changed files (bug magnets)
- whats_changed: uncommitted work in progress
- changes_to: commit history for a symbol's file

### Build System Awareness
- get_build_targets: CMake, .csproj, package.json targets with dependencies
- get_platform_variants: platform-specific code for a symbol
- platform_conditionals: all #ifdef blocks in the project

## Key facts
- 26 MCP tools
- 11 languages: Python, C, C++, C#, JavaScript, TypeScript, PHP, Dart, Swift, Kotlin, Java, Go
- 10 document formats: PDF, DOCX, XLSX, HTML, CSV, email, images, text, RST, Markdown
- 4 embedding providers: Ollama (local), OpenAI-compatible, Cohere, Voyage
- Fully local, zero cloud dependencies
- GPU-accelerated vector cache (~3ms queries)
- Multi-repo workspaces (SQLite ATTACH+UNION)
- MIT license

## Links
- GitHub: https://github.com/srclight/srclight
- PyPI: https://pypi.org/project/srclight/
- Website: https://srclight.dev