Designers
You already live in Figma
Wireframes, screens, flows, systems. Output as frames — not a PNG of a hallucination. Skills are reviewed before they hit the catalog.
You use this when “the AI designed it” is not a compliment.
Curated catalog
Skills that
aren’t slop.
Design skills for Figma, Claude, Cursor, and ChatGPT — and anyone who talks to an agent without living in a terminal. Every skill is reviewed. The agent follows a contract. You get files you can keep.
Design Engineering Marketing No terminal required
100 in the catalog 61 on GitHub 31 on Figma, npm & catalogs
Wherever the team already talks to an agent.
Cursor if you ship code. Claude or GPT if you don’t. Figma if the file is the product. Same curated catalog — not a different app for each role.
MCP host
Cursor
The agent already sits in the repo. Skills land next to the code — not in a separate design tab.
- You get
- Quote in the chat. Brand packs in .skills/exports/. UX/UI skills return a contract the agent (or Figma plugin) can place. Same five tools.
- Why it is worth it
- One MCP config. Switch models inside Cursor — logo, screen, or flow still comes from the same hosted skill, not from whichever LLM you picked today.
- How you connect
- Paste the server once. Agent discovers the catalog. No skill folders to copy.
"mcpServers": {
"design-skills": { "url": "https://mcp.design-skills.com" }
} Who it’s for
One registry. Four doors in. You just need an agent and work that shouldn’t look generated.
Developers
You already live in the repo
One contract in context — not 200 markdown files. Files land next to the code. The skill updates without a zip.
You use this when the agent should call a skill, not invent a brand kit from memory.
Marketing & growth
You need a deck on Thursday
Brand packs, packshots, editorials, landing layouts. Named artifacts. No three-week studio cycle, no Midjourney-in-Slack as the brand.
You use this when the campaign can’t look like everyone else’s prompt.
Everyone else
You don’t write code. That’s fine.
Claude, ChatGPT, Figma’s agent — same catalog the engineers use. No terminal, no bun install.
You use this when “just ask the chatbot” keeps coming back as mush.
Not a dump of 200 generic agent skills. Not unreviewed prompts. Curated brand, UX/UI, and Figma — on purpose.
Curated. Contracted. Keepable.
We don’t publish a prompt and hope. Each skill lists inputs and artifacts. Review before it ships. That’s why it doesn’t look like everyone else’s ChatGPT tab.
Four different marks
SVG, favicon, lockup rules, and a written rationale — not four recolors of one owl.
A kit in git
Signature, color, type, voice, tokens. A working brand book — not a moodboard in chat.
Not a prompt dump
The page already said what comes back. The agent can’t invent a mystery pack and call it identity.
Files you keep
Named artifacts in the repo. When the skill updates, every connected agent gets v0.2 — no stale zip.
How it works
Paste a URL or click Connect. The agent loads one reviewed skill — not a pile of prompts. You don’t have to be the person who configured it.
-
Connect
Paste the MCP URL in Cursor, Claude, GPT, or Figma. Once. Catalog is public. No skill folders to copy.
https://mcp.design-skills.com -
Pick one skill
The agent lists slugs, then loads get_skill_docs for a single contract — not 200 prompts.
-
Quote, then run
Free skills go. Premium skills show credits first. You approve. Then the hosted worker starts.
-
Output + receipt
Worker finishes in the cloud. Agent writes the pack to .skills/exports/ (the server cannot). Figma skills: plugin draws. Paid completed run pays the author.
The agent’s five tools
- list_skillsshort catalog
- get_skill_docsone contract
- quote_skillprice before run
- run_skilljob + run-id
- get_run_statusdeliverTo + files
What you get vs what stays
Comes back
Files (SVG, PNG, decks) and/or Figma structure (frames, variables, components, flows). Plus notes, manifest, receipt — exactly what the skill page listed.
Never leaves
Skill source, model keys, other people’s skills in context, a card form on the public page.
Uses and ratings use the same completed-run counter. Pageviews and quotes do not count.
Full walkthrough →You stay in control.
Clear contracts. Approval before paid work. Full audit trail.
Public skills
Browse the registry. Each skill has a name, price, and output contract.
Approval gates
Paid or sensitive skills pause before running.
Logs
Every run keeps steps, outputs, and receipts.
Artifacts
Files in the workspace and/or frames on the Figma canvas. Source of the skill stays remote.
Start with these
Same contracts the agent lists from the registry. No fake weekly-use theater.
Loading catalog…
A short shelf. On purpose.
The most-installed ones — one per origin. GitHub, Figma Community, and other catalogs: linked, not hosted. Browse the catalog →
Loading catalog…
Notes from the shelf.
Short essays on skills, slop, and why the catalog stays small.
Why agent design looks like slop
Agents fail at taste because nobody gave them a contract — only a vibe.
What a design skill actually is
A reviewed brief the agent can load. Named inputs. Named files.
A curated catalog is not a prompt dump
Stars are not review. A short shelf beats 200 generic prompts.