Hi Product Hunt 👋 I built Madar because AI coding agents often spend too much time rediscovering the same codebase. They grep, read files, summarize, lose context, then repeat the same exploration in the next task. Madar tries to make that workflow more disciplined. It builds a local structural graph of your TypeScript/Node.js workspace and compiles compact, task-specific context packs for AI coding agents. It works with Claude Code, Codex, Cursor, Copilot, Gemini, Aider, and OpenCode. The project was previously published as graphify-ts, but it has now moved to @lubab/madar with the new `madar` CLI. The latest release includes: * local graph generation * MCP integrations * context-pack-first workflows * graph summary * execution slices * benchmark commands * first-pass routing-controllers support * share-safe benchmark reports In real backend benchmarks, Madar has reduced provider-reported input tokens significantly for some prompts. This is not a universal guarantee, but it shows that better context preparation can reduce token waste and improve agent focus. I would love feedback from developers using AI coding agents, especially around MCP workflows, token waste, codebase context, and benchmark design. Thanks for checking it out.

Madar

Compact codebase context for AI coding agents

Compact codebase context for AI coding agents