Every AI coding tool wants its own context file. Claude Code reads CLAUDE.md. Cursor wants .cursorrules. Codex looks for codex.md. Windsurf needs .windsurfrules. If you're using more than one of these tools - and most serious developers are - you're maintaining multiple files that say roughly the same thing in slightly different formats.
What Happened
A developer launched codebase-md on Hacker News on March 6, 2026. The open-source tool scans a repository once and generates context files for all major AI coding assistants from a single source. It reads your project's architecture, conventions, dependencies, and file structure, then outputs the appropriate format for each tool.
The project is available on GitHub and handles the creation of CLAUDE.md, .cursorrules, codex.md, .windsurfrules, and AGENTS.md. One scan, multiple outputs.
Why It Matters
Context files are the single biggest lever you have for improving AI coding assistant output. A well-written CLAUDE.md or .cursorrules file turns vague, generic code suggestions into responses that match your project's patterns, naming conventions, and architecture. The difference between a blank context and a good one is significant - fewer corrections, less back-and-forth, and code that actually fits your codebase.
But these files rot fast. You add a new dependency, restructure a directory, change a convention - and suddenly your context file is lying to the AI. Most people write them once and forget about them. Multiplying that maintenance burden across three or four different tool formats means it basically doesn't happen.
Having a tool that reads the actual codebase means the context stays accurate. Run it again after a refactor and you're current across all your AI tools simultaneously.
Our Take
This solves a real pain point, but with caveats. Auto-generated context files tend to be heavy on structure and light on the things that actually matter most: project-specific rules, patterns to avoid, and workflow instructions that only a human would know. We maintain a 200+ line CLAUDE.md by hand for exactly this reason - the valuable parts aren't things you can extract from a directory scan.
The sweet spot is probably using codebase-md as a starting template, then layering your own rules on top. That's still a significant time savings over writing the boilerplate from scratch for each tool format.
Worth watching: as AI coding tools converge on something like a standard context format, tools like this become more or less necessary depending on whether the ecosystem standardizes or keeps fragmenting. Right now, fragmentation is winning, which makes codebase-md genuinely useful.
If you're actively using Claude Code and Cursor on the same projects, this is worth trying today.