Docs

Use CloneKit in any AI

Connect once. Then your AI can find, read and run kits of expert know-how.

01What CloneKit is

A kit is how an expert does a job, written for AI. A read kit is files your AI reads. A run kit is steps your AI follows with your tools and accounts. Then it checks the work against the expert's checks.

  • Your AI does the work. CloneKit gives it the kit.
  • Your AI always gets the latest version.
  • Anyone can search. Running, buying and publishing need an account.

02Connect your AI

CloneKit is one MCP server for every kit. Add this link to your AI app. The first time, sign in to CloneKit and tap Allow.

MCP server link

https://clonekit.ai/mcp
  • Claude

    Open Settings → Connectors and choose “Add custom connector”. Paste the link and sign in to CloneKit (tap Allow).

    Open settings
  • ChatGPT

    Open Settings → Apps & Connectors → Advanced and turn on Developer mode. Create a connector with the link and sign in to CloneKit (tap Allow).

    Open settings
  • Grok

    Open Connectors and choose New Connector → Custom. Paste the link and sign in to CloneKit (tap Allow).

    Open settings
  • Cursor

    Tap the button: Cursor opens and asks to install CloneKit. Sign in to CloneKit when Cursor asks.

    Add to Cursor
  • VS Code

    Tap the button: VS Code opens and asks to install CloneKit. Sign in to CloneKit when it asks.

    Add to VS Code
  • Claude Code

    Run this in your terminal, then start Claude Code.

    claude mcp add --transport http clonekit https://clonekit.ai/mcp
  • Codex

    Run this in your terminal, then start Codex.

    codex mcp add clonekit --url https://clonekit.ai/mcp
  • Any other AI

    Run this in your terminal and pick your apps (Windsurf, Gemini CLI, Zed, Goose and more).

    npx add-mcp https://clonekit.ai/mcp --name clonekit

Then ask your AI: “Find a CloneKit kit that helps me write a project brief.”

App without sign-in? Make an API key in Dashboard → Connections and send it as Authorization: Bearer ck_….

Just browsing? https://clonekit.ai/mcp/read is read-only: search kits and read free kits, no sign-in.

03CLI

The CLI finds, makes and publishes kits from your terminal. It needs Node.js 18 or later. Run it with npx, or install it once.

  1. Connect CloneKit to your AI apps

    npx clonekit-cli setup

  2. Sign in

    npx clonekit-cli login

  3. Find a kit

    npx clonekit-cli search "landing page design"

  4. Or install it once. Then type clonekit

    npm install -g clonekit-cli

  5. See every command

    npx clonekit-cli --help

04Skills

Kits and Agent Skills (SKILL.md folders) work together. Turn a skill into a kit, or a kit into a skill.

Get a kit as a skill

Any free read kit downloads as a SKILL.md folder:

npx clonekit-cli export @creator/kit-name

Then copy the folder to ~/.claude/skills/ or your agent's skills folder.

Turn a skill into a kit

Have a SKILL.md folder, an AGENTS.md or a public GitHub repo? Import it as a draft kit. You can also paste a GitHub link in the chat and ask your AI to import it.

npx clonekit-cli import ./my-skill

05Make and publish a kit

With your AI

Connect CloneKit, start a new chat and say:

Interview me and write my CloneKit kit.

Your AI asks one question at a time and shows you a preview before it publishes.

By hand

Write kit.yaml and your files, check them, then publish. You see a preview first.

npx clonekit-cli init my-kit
npx clonekit-cli validate ./my-kit
npx clonekit-cli publish ./my-kit

Checks before a kit goes live

  • Every kit gets an automatic safety check. Free kits go live when they pass.
  • Paid kits also need your test run and a deeper check by agents.
  • If something needs a fix, you see what and why. Fix it and publish again.

Who can see it

public
In search and on /kits. Anyone can use it.
unlisted
Anyone with the link can use it. Not in search or on /kits.
private
Only you, or your organization's members.

To sell a kit, you need Pro or Max. See pricing.

06kit.yaml fields

A kit is a folder with a kit.yaml and its files. You need id, title, listing and some content: files your AI reads (like README.md), build steps it runs, or both.

A minimal kit

id: "@your-handle/project-brief"
title: "Write a project brief"
listing: "A one-page project brief from your notes: the goal, the audience and the next action."

# Put the know-how in README.md, next to kit.yaml.

Add steps to make it a run kit

deliverables: ["A project brief"]
build:
  - id: write-brief
    step: "Write the brief"
    instructions: "Use the user's notes. State the goal, audience and next action."
    check: "The brief includes all three."
review:
  - "The brief matches the user's notes."

Required

id
The kit's name: "@your-handle/kit-name". Lowercase letters, numbers and dashes.
title
The name people see. 3 to 80 characters.
listing
What a person gets, in plain words, in about 200 characters (20 to 400). Write {tool-id} or {ref-id} only for a tool or ref the kit really uses: it shows as a chip with its logo. No chips is fine. No links, no promised results.

About the kit

about
Optional, for AI agents and search: who the kit is for, when to use it, when not to, and what it needs. Up to 1,000 characters. Cards don't show it.
version
Like "1.2.0". Starts at 1.0.0. Raise it to publish a new version.
mode
read or run. Optional: build steps make a run kit, no steps make a read kit.
domain
marketing, design, engineering, operations, sales, content, finance or other (the default).
language
The language the kit is written in: en (the default) or es.
i18n
A translated title and listing per language, like i18n: {es: {title, listing}}. Use the same {ids} as the listing.

Tools and references

tools
Apps the kit uses. Each has an id and a url, plus optional name, mcp (an https MCP link or an MCP Registry id, never a command), required (true by default), alternatives (other tool ids), recommended and est_cost.
refs
Skills, sources and templates the kit uses: id, name and kind (skill, source or template), plus optional url and note.
examples
Real results, shown on the kit's page: type (image, video, audio, before_after or doc), src under examples/ and a caption. Optional: poster, before, input and aspect.

Run steps

understand
Questions your AI asks first: ask, plus optional id, why and changes.
choose
Choices the user makes: id, question and 2 to 10 options (id, label, and optional asset and note).
build
The steps: id, step (a short title) and instructions (text or a file path). Optional: tool, check, and approval: true for steps with real effects.
deliverables
What a run kit produces. A run kit needs at least one.
review
The final checks your AI makes before it finishes.

What the user brings and gets

inputs
What the user needs to have.
outputs
What the user gets.
guardrails
Rules your AI keeps the whole time.

Price and versions

pricing
model: free (the default) or one_time with amount_usd. client_use_usd adds a higher price for client work.
license
own_use (the default) or client_use.
refund
Required for paid kits: 7_days, deliverables_not_produced (run kits) or none.
changelog
What changed. One line, or a list of versions, newest first: version and summary, plus optional date and changes. Each new version needs an entry.

Links go only in tools[].url and refs[].url. Other kit text can't hold links.

Check a kit folder before you publish:

npx clonekit-cli validate ./my-kit