Skip to content

BeBraveBeKind/mcpskills

Repository files navigation

mcpskills

Trust Score License: MIT npm

Trust-score any AI skill or MCP server. 14 signals across 4 dimensions with safety scanning for prompt injection, credential theft, and supply chain attacks.

Website: mcpskills.io

How It Works

mcpskills scores GitHub repos across 4 dimensions:

  • Alive — Is it maintained? (commit recency, release cadence, issue responsiveness)
  • Legit — Is the author credible? (author credibility, community adoption, contributor diversity)
  • Solid — Is it secure? (security posture, dependency health, tool safety, supply chain safety)
  • Usable — Is it well documented? (README quality, spec compliance, license clarity)

AI skills and MCP servers get enhanced scanning with 5 safety checks based on ClawHavoc and ToxicSkills attack patterns.

Quick Start

MCP Server (Claude Code, Cursor, Claude Desktop)

Install the MCP server to score repos directly from your IDE:

claude mcp add mcpskills -- npx @mcpskillsio/server

See mcp-server/README.md for Cursor and Claude Desktop setup.

Website

Scan any repo at mcpskills.io — free, no signup required.

API

curl -X POST https://mcpskills.io/api/score \
  -H "Content-Type: application/json" \
  -d '{"repo": "anthropics/anthropic-sdk-typescript"}'

Architecture

mcpskills/
  mcp-server/          # npm package (@mcpskillsio/server) — 8 MCP tools
  lib/                 # Shared core: scorer (14 signals), skills detector, safety scanner
  netlify/functions/   # Serverless API (score, badge, monitor, certify, webhook)
  public/              # Static website (mcpskills.io)
  data/                # Registry, score cache, curated packages
  scripts/             # CLI utilities
Component Purpose
MCP Server IDE integration — score repos from Claude Code, Cursor, or any MCP client
API REST endpoints for scoring, badges, monitoring, certification
Website Live scanner, blog, trust badge generator
Scheduled Functions Nightly crawl (2am UTC), daily monitoring (8am UTC), weekly digest (Sunday 6pm UTC)

MCP Tools

Tool Description
check_trust_score Score any GitHub repo (0-10, 4 dimensions, 14 signals)
scan_safety Focused safety scan for AI skills (5 threat categories)
list_packages Browse curated, pre-scored skill packages
get_badge Generate SVG trust badge for READMEs
watch_repo Monitor repos for trust score changes
check_watched Re-scan all watched repos
batch_check Score up to 5 repos in one call
auto_gate Boolean go/no-go decision with reasoning

Trust Tiers

Tier Score Meaning
Verified >= 7.0 High confidence across all dimensions
Established >= 4.5 Moderate confidence, sufficient signals
New < 4.5 Insufficient data or low scores
Blocked Disqualifiers: no license, critical CVE, dangerous workflows

API Key

Free tier returns trust tier + dimension scores. For full 14-signal reports with safety findings:

export MCPSKILLS_API_KEY=your_key_here

Get your key at mcpskills.io/api.

License

MIT — Built by Michael Browne at Rise Above Partners.

About

MCP server for skill trust scoring, safety scanning, and package watching

Topics

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Packages

 
 
 

Contributors