About AgentArmor
AgentArmor is a security check-up for personal AI agents. We help people who aren't security experts understand an add-on's risks before they install it.
What we check
Skills are packaged instructions and tools that teach an AI assistant a new ability. App connectors let it reach into other apps. MCP (Model Context Protocol) is the standard way AI assistants like Claude connect to other apps.
Every day we look at GitHub, the npm package library, the official MCP registry and Hacker News to find the 500 most popular items.
Popularity score
Popularity score (out of 100): combines public adoption signals such as GitHub stars and weekly npm downloads (each scaled so one huge number does not dominate), relative to the most popular item in our list, which scores 100. It measures attention, not risk.
Risk grades
- A — Few risk signals: Strong public adoption and transparency signals.
- B — Some risk signals: Minor concerns worth a quick check.
- C — Moderate risk signals: Several concerns worth a closer look before you install.
- D — High risk signals: Weak public adoption or transparency signals; review carefully first.
- F — Many risk signals: Very few public adoption or transparency signals; review it fully before you rely on it.
Based on public adoption and transparency signals. Our opinion, not a code audit or security guarantee. Each grade shows the date it was last checked.
Grades come from fixed rules applied to public adoption and transparency signals. Write-ups are drafted by an AI model (Anthropic's Claude Haiku), may contain errors, and aren't part of the grade. Spot an error? Tell us at our contact page.