The one prompt that installs all four
Open Claude Code, paste this, and it installs three of the four for you and reminds you to type the fourth. It reports back in plain English after each one, so you always know what just happened on your computer.
Before you paste
You need Claude Code open — Mac, Windows or Linux all work. Any folder is fine, but use the project you actually work in. It takes about ten minutes. Claude asks permission before each install, so say yes. One of them (21st) needs a free account, and the prompt stops and tells you when it gets there instead of guessing.
Then type this inside Claude Code
Anthropic's official plugin installs from the plugin menu, not from the terminal. One line, hit enter, done. The claude-plugins-official marketplace is built in, so there is nothing to add first.
Only want one or two?
Paste the big prompt anyway and delete the numbered steps you do not want before you hit enter. Claude only installs what is left. If you are only doing one, make it Humanizer — it is the fastest to install and the difference is immediate.
Which one fixes your problem
Generic AI output is not one problem, it is four. The writing sounds like a chatbot. The design looks like every other AI site. The components get invented instead of used. And the project was never set up properly in the first place. Find yours below.
"It writes like a bot"
Humanizer. Marks every AI tell in your text and rewrites it, holding your facts fixed. Paste a few paragraphs of your own writing and it copies your rhythm instead of its own.
"Everything it designs looks the same"
Impeccable. Runs 61 rules over your interface that catch the tells — purple-to-blue gradients, bounce easing, dark glows — and gives you the words to ask for something better.
"It invents components badly"
21st MCP. A searchable catalogue of real React and Tailwind components. Claude picks one that exists and installs it with its dependencies instead of writing a wonky one from scratch.
"I never set this project up"
Claude Code Setup. Anthropic's own plugin reads your project and tells you which hooks, skills, MCP servers and sub-agents would actually help here. It only recommends, so nothing can break.
The prompt installs all four. If you would rather go one at a time: Humanizer first because you will feel it immediately, then Claude Code Setup so you know what your project is missing, then Impeccable, then 21st the next time you are actually building an interface.
The four, one at a time
What each one actually does, a prompt to prove it is working, and the catch. Strictly speaking two are skills, one is a plugin and one is an MCP server. I am calling them all plugins because that is what they feel like.
Humanizer · strips the tells that make writing sound like AI
It reads your text, marks every AI tell it finds worst-first, then rewrites without touching your facts. The patterns come from Wikipedia's own field guide to spotting AI writing, grouped into five habits: staging instead of stating, rhythm by rule, inflated and borrowed authority, formatting by rule, and leftovers from the chat window.
The part most people miss is the voice matching. Paste two or three paragraphs of your own writing with the request and it follows your rhythm, your word choice, your punctuation and your deliberate quirks — dashes included, if you use them.
The catch: the pattern count keeps moving. It was 33, then 35, and as of 6 September 2026 it was rebuilt down to 25 patterns ordered by strength. So do not quote a number at anyone. Several patterns are tagged "weak alone" — em dashes, passive voice — because careful humans use them on purpose, so it will not strip yours if your writing sample uses them. It also refuses to invent: if a sentence needs a fact that is not there, it stops and asks you instead of smoothing over the gap. See it on GitHub →
Impeccable · catches the design cliches
It installs 23 slash commands that give you and Claude the same design vocabulary — /impeccable polish, audit, critique, bolder, quieter — so "make it less AI-looking" becomes something Claude can actually act on. Behind those sits a set of 61 rules that run on your machine with no model and no API key, flagging the tells: the purple-to-blue gradient, the bounce easing, the dark glow, cramped padding, tiny tap targets, line lengths nobody can read.
Run /impeccable init first. It writes a PRODUCT.md recording who your product is for and what it sounds like, and every later command reads it. Skip that step and the advice stays generic.
Then point it at the page you like least:
The catch: a clean run is not proof your site looks good. The author says so himself — it "does not replace inspecting the rendered experience", it just means you have stopped making the obvious mistakes. It also works by subtraction. It has no library of beautiful sites to copy, it only knows what to stop doing. And the live editing mode only runs against a site on your own computer; for a published site you get a read-only check. See it on GitHub →
21st MCP · real components instead of invented ones
Ask Claude for a pricing table and it writes one from memory, which is how you end up with something that almost works. 21st gives it a catalogue to search instead — over 12,000 React components, marketing blocks, templates and shadcn themes — and it installs the one you pick with its dependencies, into your project, matching your existing setup.
It used to be called Magic MCP. That package still exists but it is now just a redirect, so install it with the 21st CLI line in the prompt above and ignore any tutorial older than this year.
The catch: this is the one with a real limit. Searching and browsing all 12,000 is free, but the free tier lets you actually copy two components a day. After that you are paying. The AI generation feature is a separate paid credit add-on on top of that. Two a day is genuinely enough to build something over a week, but do not plan a whole site for a Sunday afternoon. Old Magic API keys were also reset, so if you set this up before this year, get a new one. See 21st.dev →
Claude Code Setup · Anthropic's own setup check
This one is made by the people who make Claude Code. It reads your project — your dependency files, your folder structure, your existing CLAUDE.md and test config — and comes back with the top one or two things worth adding in each category: MCP servers, skills, hooks, sub-agents and slash commands. Instead of installing whatever went viral this week, you find out what your project is actually short of.
After install nothing looks different. The change is that you can ask Claude what to set up instead of guessing.
The catch: it reads, it does not build. Anthropic's own documentation says it plainly — the skill is read-only, it analyses and outputs recommendations and it creates or modifies nothing. You still say yes to each suggestion and get Claude to set it up in a separate step. It also holds itself to one or two ideas per category so you are not buried. If you want more, ask for one type at a time — "what hooks should I use for this repo?" gets you a longer list. See the plugin on GitHub →
Take it all out again
None of this is permanent. If you want to start clean, paste this and Claude removes all four and puts your settings back.
Keep whatever solved a real problem and bin the rest. Most people end up keeping Humanizer and Claude Code Setup and dropping the other two once the novelty wears off. That is a perfectly good place to land.
Ready to go deeper?
Where to go from here.
Wright Mode Membership
Join a community of women entrepreneurs implementing AI and automation in their businesses. Live calls, templates, and ongoing support.
Claude Masterclass
Learn how to use Claude like a pro. From prompting fundamentals to building real workflows that save hours every week.
Claude Code Masterclass
Go beyond the chat interface. Build automations, process data, and create tools with Claude Code — no developer experience needed.