Methodology
Claudius ranks Claude plugins, MCP servers, skills, subagents, hooks and slash commands using public, verifiable signals. Data refreshes daily via an automated pipeline; metric numbers are pulled directly from the GitHub, npm, PyPI and OSV.dev APIs and are never hand-edited.
Default ranking
A composite score, weighted:
- 40% — popularity (log-scaled GitHub stars)
- 30% — trust score (below)
- 20% — growth (log-scaled 30-day star delta)
- 10% — recency of last commit
Every column is sortable if you prefer a single signal.
Trust score (0–100)
Starts at 50, then:
- +20 official / first-party, +10 established community org
- +15 commit in last 30 days · +8 within 180 days · −15 dormant >1 year
- +8 OSI license · −5 no license
- up to +10 social proof (stars, capped at 10k)
- +5 repo older than a year
- −25 per known security advisory (OSV.dev / GitHub Advisory DB, capped at −40)
Grades: A ≥ 85 · B ≥ 70 · C ≥ 55 · D ≥ 40 · F < 40. A trust score is a heuristic, not an audit — always review what you install. Extensions execute with your permissions.
Discovery & curation
New entries are discovered daily from the official MCP registry, GitHub topics, npm and community marketplaces, then reviewed by an automated curation pass that filters forks, mirrors, name-squats and non-extensions. Precision over recall: uncertain candidates are rejected. Delisted repos are flagged, never silently dropped.