AgentStore
Methodology

How we score and rank

Fully public. If you're a tool author, this tells you how to "game" us — by design, because we assume you will read it.

01

AgentStore Score formula

AgentStore Score =
0.40 × install_score + 0.30 × activity_score + 0.30 × rating_score

Recalculated every Saturday at 00:00 UTC.

02

How each component works

Install score (40%)

Actual usage is the most important. We take Smithery calls, npm / PyPI monthly downloads, then normalize.

install_score = log10(smithery_calls + npm_monthly × 3 + pypi_monthly × 3) / 6

npm / PyPI monthly × 3 to be comparable to cumulative Smithery calls. log10 smooths the long tail.

Activity score (30%)

Measures maintenance and community engagement.

activity_score = (commit_recency + contributors + issue_response) / 3
  • · commit_recency: days since last commit, <30d=1.0, >180d=0.0
  • · contributors: log10(# contributors) / 2, max 1.0
  • · issue_response: median response <48h=1.0
Rating score (30%)

Weighted blend of user reviews and our editors' assessment.

rating_score = user_avg × min(count / 10, 1.0) + editor_score × (1 - min(count / 10, 1.0))

Editor weight dominates below 10 reviews; user ratings take over above that threshold.

03

The 6 rankings

Overall
All tools sorted by AgentStore Score
Weekly growth
Past 7 days growth (Smithery + npm), ≥30 days of data required
Rising stars
Launched ≤60 days + cumulative calls ≥1,000
Underrated
GitHub stars <100 + Smithery calls ≥1,000
Overrated
GitHub stars >200 + Smithery calls <500
Dead
Last commit >180 days ago, or repo archived
04

Anti-gaming measures

  • · npm / PyPI downloads filter out CI patterns (regular scheduled pulls don't count)
  • · Same-IP burst Smithery calls are rate-limited before counting
  • · Anomalies get flagged; proven gaming leads to deranking or removal
05

How our own tools are handled

AgentStore Studio tools use the same algorithm. They display an "AgentStore Studio" badge for transparency, but receive no score bonus. If an in-house tool tops a ranking, we'll explain why in the data.

06

Disagree with a ranking?

Open a GitHub issue, or email methodology@agentstore.xyz with the data you're seeing. We respond publicly.