Case study
cgb - Code Graph Builder
Open-source CLI, library and MCP server that maps how a codebase connects and gives AI coding assistants small, relevant context bundles.
Overview
AI coding assistants waste a large share of every session reading whole files just to understand how code fits together — the cgb README estimates 50–100K tokens per session. cgb builds a persistent graph of a codebase (dependencies, calls, inheritance) and turns it into compact Markdown context bundles of roughly 1,000–5,000 tokens that contain exactly what the assistant needs.
Key features
- Dependency and impact analysis: see what a change will touch before you make it
- Context bundles sized for AI assistants
- Git change detection with risk scoring and review context
- Callers, code communities and dead-code detection
- Wiki generation and embedding-based semantic search
- Watch mode, an MCP server (with read-only mode) and a VS Code extension
- Supports 11 languages: TypeScript, JavaScript, Python, C#, Java, Go, Rust, Ruby, PHP, C/C++ and Kotlin
Tech stack
- TypeScript on Node.js 20+
- Tree-sitter (WASM) parsers for multi-language analysis
- better-sqlite3 for the persistent graph
- Model Context Protocol SDK and chokidar file watching
Get it
MIT licensed. Install from npm as cgb-builder or browse the source on GitHub.