Simon Willison Ships 66MB SQLite DB du navigateur MDN Compact Data avec Open CORS

Simon Willison just published a 66MB SQLite database containing the entire mdn/browser-compat-data repository, accessible via GitHub's CDN with open CORS headers.

He built it after Mozilla launched their MDN MCP service and realized the underlying browser-compat-data repo was ripe for local querying. The conversion script was generated by Claude Code for web (Opus 4.8) using sqlite-utils. The actual GitHub Actions workflow? That came from Codex Desktop (GPT-5.5).

Why Bother with an Orphan Branch

GitHub Releases don't serve files with open CORS headers. Regular repository files do. Willison wanted anyone to download or explore the database directly from Datasette Lite without a proxy. So he made the Actions workflow build the SQLite database and force-push it to a standalone orphan branch named "db".

The database sits at a predictable URL under his user.github.io-style raw path, with CORS headers intact. You can grab it here or open it instantly in Datasette Lite.

The Result: Instantly Queryable Browser Data

Any developer can now run SQL queries against every CSS property, JavaScript API, and HTML element compatibility table from MDN without hitting a browser or parsing JSON. The schema mirrors the original tree structure, flattened into tables.

This is the kind of data plumbing that usually gets ignored until you need to answer "does Safari support backdrop-filter in 2023?" across a fleet of test runners. Now that answer is a single SELECT away.

A Pattern Worth Stealing

Hosting a structured dataset on an orphan branch is trivial to set up and costs nothing beyond the repo storage. Willison's combo of AI-generated scripts and a dead-simple GitHub Actions pattern lowers the barrier for anyone who wants to ship a SQLite database with open CORS. Expect to see more 10MB+ data files showing up as orphan branches in the coming months.

Source: simonw/browser-compat-db
Domain: simonwillison.net