Agents

Jarvis UI is a shadcn registry, so the tools shadcn makes for coding agents work with it too. And every page of these docs is also served as Markdown, for agents to read.

shadcn's tools

Installation adds the @jarvis registry to your project's components.json. shadcn's tools read the registries listed there, so an agent can find and add the components by name, as @jarvis/dialog, without being told where the registry is.

Skill

shadcn's skill teaches an agent the shadcn CLI and how a shadcn project fits together. Add it to your project.

The agent loads it whenever it works in a project with a components.json. It runs shadcn info to learn the project's framework, aliases and installed components, and looks components up before writing code with them, rather than guessing at their APIs.

MCP server

shadcn's MCP server gives an agent tools to browse, search and add items from the project's registries. Set it up for Claude Code.

--client also takes cursor, vscode, codex and opencode. Then ask for components in plain language:

• "Show me the components in the @jarvis registry"
• "Add @jarvis/dialog and @jarvis/form"
• "Build a settings page with a side nav, using components from @jarvis"

CLI

Agents can run the CLI themselves too. search finds items in a registry, and view shows an item's files and dependencies before it's added.

shadcn docs only knows shadcn/ui's own components, though, and many of them share a name with one of these. shadcn docs button describes shadcn/ui's Button, whose variants and props differ from Jarvis UI's, so point agents at the Markdown below instead.

Docs as Markdown

Every docs page is also Markdown, at its URL plus .md: /components/button.md, /docs/theming.md, and /docs.md for the Introduction. Each example is its source, each props table a Markdown table, and each command is written for npm. A component's page links to its source on GitHub as well.

• /llms.txt indexes every page with its description, in the format llmstxt.org proposes.
• /llms-full.txt has every page in one file, to give an agent all of the docs at once.
• /components.md lists the component pages by group, as the gallery does.

Agents don't need to know the addresses. A request that accepts text/markdown gets the Markdown at a page's own URL.

curl -H "Accept: text/markdown" https://ui.jarv.is/components/button

Each page also links to its Markdown with <link rel="alternate" type="text/markdown">, and to /llms.txt with <link rel="describedby">, which the responses repeat as Link headers.

The Copy page button at the top of each page copies its Markdown. Its menu opens the Markdown, or starts a chat in v0, ChatGPT, Claude, Cursor or Grok that begins by reading the page.

Registry items

A component's page answers the shadcn CLI with its registry item, so an agent holding a link to the docs can add the component from it.

/r/registry.json lists every item, with its files and dependencies.

Telling your agent

Agents read AGENTS.md (or CLAUDE.md, for Claude Code) at the start of each session. A few lines there send them to the right docs.

## UI components

The components in `components/ui` come from Jarvis UI, a shadcn registry. Before using one, read
its docs at https://ui.jarv.is/components/<name>.md rather than shadcn/ui's, whose components of
the same name differ. Find more at https://ui.jarv.is/llms.txt, and add them with
`npx shadcn@latest add @jarvis/<name>`.