MCP servers give an agent tools. Skills tell it how to use them.
For a while now, the interesting work in the Claude Code ecosystem hasn't only been new servers — it's been the growing library of skills and plugins that change how an agent plans, writes, tests, and ships. We kept getting asked where to find them. So we built a page.
/skills is live today: 65 hand-picked Claude Code skills, plugins, and marketplaces across 10 categories, searchable and filterable, with install instructions for every agent we could verify.
What's in the directory
The day-one lineup includes the skills most people have already heard of and several that deserve more attention:
- Superpowers — Jesse Vincent's agentic development methodology: brainstorm → spec → plan → TDD → subagent execution → review → merge, wired up as auto-triggering skills.
- Anthropic Agent Skills — Anthropic's official open-source library (canvas design, DOCX/PDF/PPTX/XLSX generators, frontend design, MCP builder, skill creator, and more).
- UI UX Pro Max — high-end interface design workflows for Claude Code.
- Awesome Agent Skills — the community-maintained index, re-surfaced inside MCPpedia.
- MCP Builder, Skill Creator, Webapp Testing, Playwright Skill, claude-mem, Continuous Claude v3, Planning With Files, and dozens more.
Categories cover the full range of what skills are actually being built for: development workflow (18), meta/collections (14), data (9), design (7), writing, research, testing, product, marketing, and a catch-all for the genuinely weird.
What we filtered out
GitHub's topic:claude-skills tag is noisier than it looks. A lot of repos slap the tag on for visibility without shipping an actual skill. The starting filter was topic:claude-skills stars:>50, and then we went through the list by hand and cut anything that wasn't a real, installable skill.
The number we cared about wasn't coverage. It was signal-to-noise on the first page.
The directory is curated, not crawled. Unlike MCP servers — where bots continuously ingest everything they can find — skills are added by hand. That's deliberate for launch. If the section earns traffic, we'll graduate it to the same automated pipeline the server catalog runs on.
Install paths for every agent we could verify
Each skill page shows the install command for whichever agents it supports — Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, OpenCode. Where the install is manual, we say so and link the README instead of pretending there's a one-liner.
A skill's detail page also surfaces its author, license (where declared), GitHub stars, last-updated date, and related skills in the same category. Everything is linkable and SEO-indexed — SoftwareApplication JSON-LD on each skill, CollectionPage + ItemList on the index.
What's next
The shortlist of follow-ups:
- Submissions. Right now, additions happen via PR against
data/skills.json. A proper submission form is next. - Scoring. MCP servers get an 8-check safety/quality score. Skills probably need a different rubric — install cleanliness, skill metadata quality, README signal — but the principle (show the evidence, not just the number) carries over.
- Automation. Once the submission flow exists, the bots can help keep stars, last-updated dates, and broken links fresh.
If you maintain a skill that should be on the list, or if you spot something miscategorized, open an issue. Coverage is the goal — but only for things that are actually skills.
Start browsing: mcppedia.com/skills.
MCP Security Weekly
Weekly CVE alerts, new server roundups, and MCP ecosystem insights. Free.
Keep reading
This article was written by AI, powered by Claude and real-time MCPpedia data. All facts and figures are sourced from our database — but AI can make mistakes. If something looks off, let us know.