AgentSkillsHub — published at agentskillshub.top — is an open-source, security-graded directory of Claude Skills, MCP servers, Codex skills and AI agent tools that indexes more than 267,000 GitHub projects and refreshes every eight hours, built for developers working in Claude Code, Cursor, Windsurf, Cline, OpenAI Codex or any other agentic IDE who need to find a trustworthy component without searching GitHub topic pages by hand.

What it is

AgentSkillsHub is an MIT-licensed directory project, written primarily in HTML, that continuously collects open-source AI agent skills, MCP servers, Codex skills, AI coding assistants and agent frameworks from GitHub and turns them into a browsable catalogue. Every project it ingests is classified into one of seven categories (mcp-server, claude-skill, codex-skill, agent-tool, prompt-library, ai-coding-assistant, uncategorized), scored on ten weighted signals and six quality dimensions, and given a security grade. The stack is a Python 3.12 / FastAPI / SQLAlchemy backend with a React 18 / TypeScript / Vite / TailwindCSS front end, backed by Supabase (PostgreSQL) in production or SQLite for local runs, and deployed to GitHub Pages through GitHub Actions on an eight-hour cycle.

The concrete problem it solves is discovery inside the agent ecosystem: the README's stated goal is that a developer should be able to find the right MCP server or Claude Skill in under 30 seconds instead of spelunking through random GitHub topics. It replaces ad hoc topic browsing and unranked keyword search with a scored, categorised index, side-by-side /compare/ views and scenario landing pages under /best/{scenario}/, and it adds a security grade on top of the usual popularity signals so that choosing a component is not purely a matter of counting stars.

Key capabilities

• Classifies every repository into one of seven fixed categories, covering MCP servers, Claude skills, Codex skills, agent tools, prompt libraries and AI coding assistants.
• Computes a composite 0–100 score from ten weighted signals, including a quality aggregate, logarithmically normalised star counts and exponential recency decay.
• Breaks quality down into six dimensions: completeness, clarity, specificity, examples, README structure and agent readiness.
• Attaches a security grade to each listed project and publishes a linked security report.
• Offers side-by-side comparison through /compare/ and per-scenario shortlists through /best/{scenario}/ pages.
• Runs a six-phase pipeline in scheduler/jobs.py — Search, Masters, Extra, Enrich, README, Upsert — with incremental weekday syncs and a full Sunday sync.
• Deduplicates and normalises records in services/data_cleaner.py by repo_full_name, infers project type, and buckets projects by size from micro (≤50KB) to large (>5MB).

Who uses it and how

• Developers working in Claude Code, Cursor, Windsurf, Cline or OpenAI Codex use it to shortlist an MCP server or skill before adding it to an agent setup.
• Reviewers comparing candidate skills can weigh quality scores and security grades side by side rather than judging repositories from their README alone.
• Contributors can submit projects, which enter the pipeline through the community-submitted and curated repositories phase of the sync.
• Teams that want the catalogue in-house can self-host it, since the whole thing runs from a scheduled GitHub Actions workflow against Supabase or a local SQLite file.
• Readers who want periodic digests can subscribe to the newsletter, delivered through the Resend API and Supabase pg_net.

Getting started

The primary route is the hosted directory at agentskillshub.top. Self-hosting means running the FastAPI backend against Supabase (PostgreSQL) or a local SQLite database with the React/Vite front end and deploying through GitHub Pages and GitHub Actions.

How it compares

The facts provided list no paid products that this project replaces, and they name no comparable skills directory. Among the tools mentioned — Claude Code, Cursor, Windsurf, Cline and Codex — those are agent clients that consume skills rather than catalogues of them, so AgentSkillsHub stands alone in this registry.

When to use it — and when not to

Self-hosting means operating a real data pipeline: a Supabase or SQLite database, GitHub API access with rate-limit handling, a Resend integration for email, and a GitHub Actions schedule that must complete within per-phase budgets. The scores are derived from repository signals such as README structure, topics, stars and commit activity, so a high grade reflects documented, well-structured projects rather than verified runtime behaviour, and the security grade is a signal to inform your own review rather than a replacement for it. Teams that want a small, human-curated list of vetted skills, or that need sources outside GitHub, should look elsewhere; the project's own repository also carries 38 open issues, so expect an actively moving target.

GitHub API ──▸ Collection ──▸ Cleaning ──▸ Evaluation ──▸ Scoring ──▸ Presentation
               (6 phases)    (classify)   (6 dimensions)  (0-100)    (Web + API)

cd backend
python3.12 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env   # Add GITHUB_TOKEN
uvicorn app.main:app --reload

cd frontend
npm install
npm run dev

┌─────────── GitHub Actions ───────────┐
│  sync.yml (8h) · deploy.yml · newsletter.yml (Mon) │
└──────┬──────────────┬────────────────┬──┘
       ▼              ▼                ▼
   Supabase      GitHub Pages     Resend API
  (PostgreSQL)    (React SPA)     (Emails)

GITHUB_TOKEN=ghp_xxx          # GitHub API (required)
SUPABASE_DB_URL=postgresql://  # Production database
RESEND_API_KEY=re_xxx          # Email service
ADMIN_TOKEN=sk-xxx             # Admin API auth
SYNC_INTERVAL_HOURS=8          # Sync frequency

[![Listed on Agent Skills Hub](https://img.shields.io/badge/Agent_Skills_Hub-listed-2ea44f)](https://agentskillshub.top)