Kokonut UI - Open source components
Open-source components your AI agent installs over MCP
48 followers
Open-source components your AI agent installs over MCP
48 followers
Open-source components your AI agent installs over MCP, led by the AI parts agents build with.
This is the 4th launch from Kokonut UI - Open source components. View more
Kokonut UI
Launching today
Kokonut UI is a free, MIT-licensed library of 100+ React components made for coding agents. Connect through MCP in Claude Code, Cursor, VS Code or Codex, ask for "a tool approval card in my chat", and your agent finds the component, installs it and edits it in your repo. Every docs page is also served as Markdown for agents.
This release adds 65 components and rebuilds 34, led by AI chat parts: tool calls, tool approval, diff review, citations and more.
Free
Launch Team / Built With
CloudflareDeploy serverless functions globally in seconds
Promoted
Hey again :D, I'm Dorian and I make Kokonut UI, a free and open-source component library.
When I was last here in June 2025, it was a set of animated components you copied and pasted. Since then the way people build has changed: you describe a screen to Claude Code or Cursor, and the agent writes it. So I rebuilt Kokonut UI for that.
How it works with your agent:
Run npx shadcn@latest mcp init --client claude (or cursor, vscode, codex). Your agent can now search and install all 100+ components over MCP.
Or paste one line: "Set up Kokonut UI in this project by following kokonutui.com/setup.md".
Every docs page is also Markdown, and a rules file tells your agent to check Kokonut UI before building a component from scratch.
What it installs: 65 new components and 34 rebuilt, led by AI chat parts: tool calls that fold into one line, tool approval, diff review, inline citations, a context meter, and a checkout that lives inside a reply. Plus app flows (slide to confirm, seat map, card field) and a new Media category.
Everything is MIT. The code lands in your repo: no account, no lock-in, and you can read every file on GitHub.
I'll be around all day and wonder: which component you'd ask your agent for first, and what the AI set is still missing!