Daily scrape of Claude Code plugin install counts from Anthropic's public plugin-stats endpoint, with history back to December 2025 recovered from Arq backups.
Search any of the 450+ plugins, toggle log scale, zoom into date ranges, and hover for exact counts. (GitHub strips JavaScript from READMEs, so the interactive version lives on GitHub Pages — click through.)
A GitHub Action fetches
https://storage.googleapis.com/claude-code-dist-86c565f3-f756-42ad-8dfa-d59b1c096819/plugin-stats/plugin-details.json
daily and commits it if it changed — Simon Willison's git-scraper
pattern. The same run
regenerates docs/data.json (one [date, installs] series per plugin,
extracted from the full git history) which feeds the chart.
The scrape only started in June 2026, but Claude Code has cached this data
locally since late December 2025 (~/.claude/plugins/install-counts-cache.json,
later renamed plugin-catalog-cache.json). Weekly snapshots of those files
were recovered from encrypted Arq 7 backups in S3 — including thawing the
relevant pack files out of Glacier — and committed with their original
fetchedAt timestamps as commit dates. The patched restore tooling that made
this possible lives at obra/arq_restore;
the recovery process is documented in RECOVERY-HANDOFF.md.
plugin-details.json— latest snapshot from the public endpointinstall-counts-cache.json,plugin-catalog-cache.json— historical local cache formats (frozen; present for the recovered history in git log)scripts/build-history.py— walks git history →docs/data.jsondocs/— the GitHub Pages chart (ECharts via CDN).github/workflows/scrape.yml— daily fetch + commit + data rebuild