Best shadcn registries

The formula, in full

• Inventory — 34%. Item count, log-scaled. A registry with 4,000 blocks is not forty times more useful than one with 100; past a few hundred items more inventory stops being what decides whether you find what you need.
• Free share — 26%. The proportion of items whose source the crawler could actually fetch. A registry that serves no probeable item endpoint scores 40 here rather than 0 or 100 — we do not know, and pretending either way would be worse than saying so.
• Upkeep — 26%. Days since the last push to the registry’s public repository. Under two weeks scores full; over a year scores 14. A registry with no public repository scores 45.
• Manifest served — 14%. Whether registry.json comes from the published site (100) or only from the repository (55). A registry you cannot install from by URL is a registry you have to work to use.

What it deliberately does not measure

Not quality, not design, not accessibility, not whether the code is any good. None of those can be established by a crawler, and a number that pretends otherwise is worse than no number. It also does not measure popularity beyond the registry’s own GitHub stars, which are shown on the registry index and carry no weight here — stars measure how many people bookmarked a repository, not how many shipped with it.

How to actually use this

Read the top of the table once to know who the serious publishers are, then stop thinking about registries. Since every item installs by its own URL, the useful move is to search for the component you need across 62,855 indexed items and let the registry be whichever one happens to have it.

Every field in this table is available as JSON at /api/registries. The ranking is computed on this page, so you can apply your own weights to the same numbers.