AI Tools

v0 provides machine-readable documentation files following the llms.txt↗︎ standard. These files help AI assistants understand the library without hallucinating APIs or patterns.

Available Files

What’s Included

llms.txt contains categorized links to:

• Guide pages (fundamentals, features, tooling, integration, and more)

• 43 headless components (Atom, Avatar, Pagination, etc.)
• 71 composables across categories

• FAQ and contributing guides

llms-full.txt contains the complete content of every documentation page, stripped of Vue components and frontmatter for cleaner LLM consumption.

SKILL.md is a compact reference optimized for AI coding assistants. It focuses on decision trees, code conventions, anti-patterns, and the composition hierarchy. Detailed API examples live in a separate REFERENCE.md that agents load on demand.

Usage

Need a ready-made agent? Paste the setup prompt into Grok Bot↗︎ — it walks connections and saves itself. Identity fields on that page also work for Claude Code and Cursor.

Install SKILL.md via skills.sh↗︎ — works with Claude Code, Codex, Cursor, Grok Build, Windsurf, and 35+ agents↗︎:

npx skills add vuetifyjs/0

Editors and CLIs

Claude Code / Grok Build — add Vuetify MCP for structured API access:

claude mcp add --transport http vuetify-mcp https://mcp.vuetifyjs.com/mcp

grok mcp add --transport http vuetify-mcp https://mcp.vuetifyjs.com/mcp

Or fetch docs directly in your session:

WebFetch https://0.vuetifyjs.com/llms-full.txt

Grok Build also loads SKILL.md (via npx skills add vuetifyjs/0 or .grok/skills/) and project rules from AGENTS.md / CLAUDE.md.

Cursor / Windsurf — add to .cursorrules or configure MCP:

@https://0.vuetifyjs.com/llms.txt

See Vuetify MCP for IDE configuration.

ChatGPT / Claude.ai — paste the URL in chat:

Read https://0.vuetifyjs.com/llms-full.txt and help me build a multi-select dropdown.

Agent Context

With MCP — always-current structured access:

• get_vuetify0_skill — Latest SKILL.md reference

• get_vuetify0_composable_list — Browse all composables

• get_vuetify0_composable_guide — Detailed composable docs

• get_vuetify0_component_list — Browse headless components

• get_vuetify0_component_guide — Component docs and examples

• get_vuetify0_exports_list — Package subpath exports

• get_vuetify0_installation_guide — Setup instructions

• get_vuetify0_package_guide — Package-specific documentation

Without MCP — fetch SKILL.md at session start:

WebFetch https://0.vuetifyjs.com/SKILL.md

What’s in SKILL.md

• Decision tree mapping needs to composables

• Code style conventions (shallowRef, function declarations)

• Anti-patterns to avoid

• Composition hierarchy

• Summary lookup table

• Links to REFERENCE.md for detailed API examples

Making Agents Actually Use v0

Docs access alone isn’t enough. An agent only looks things up when it feels uncertain — and a model trained before v0 existed doesn’t feel uncertain writing generic Vue. It will fluently hand-roll selection state, focus traps, or virtual scrolling without ever checking whether v0 provides them. The fix is to move v0 knowledge from on-demand recall to ambient context, plus a deterministic trigger.

1. Put a Surface Map in Always-Loaded Context

Add an inventory of v0 exports — every name with a one-line purpose — to a file your agent loads on every session (CLAUDE.md, .cursorrules, AGENTS.md). An agent can’t reach for createSelection if it doesn’t know the name exists; once the name is in context, fetching the full guide via MCP or SKILL.md follows naturally.

Generate it from the MCP server (get_vuetify0_exports_list) or copy the lookup table from SKILL.md, then keep a section like:

## v0 Surface Map
If a name below covers the need, use it — never hand-roll an equivalent.
- createSelection — multi/single selection with mandatory guards
- createForm — form state + field coordination
- useHotkey — keyboard shortcuts
- createVirtual — virtual scrolling for large lists
<!-- ...full export list... -->

2. Add a Deterministic Reminder Hook

Context-file instructions degrade over long sessions; hooks don’t. In Claude Code, a PreToolUse hook fires every time the agent edits a file — including edits made by subagents — and can’t be rationalized away. In .claude/settings.json:

{
  "hooks": {
    "PreToolUse": [{
      "matcher": "Edit|Write",
      "hooks": [{ "type": "command", "command": ".claude/hooks/v0-reminder.sh", "timeout": 5 }]
    }]
  }
}

And .claude/hooks/v0-reminder.sh (injects once per session, only for Vue/TS files):

#!/usr/bin/env bash
input=$(cat)
file=$(jq -r '.tool_input.file_path // empty' <<< "$input")
case "$file" in
  *.vue|*.ts|*.tsx)
    mark="${TMPDIR:-/tmp}/v0-reminder-$(jq -r '.session_id' <<< "$input")"
    if [ ! -e "$mark" ]; then
      touch "$mark"
      printf '%s' '{"hookSpecificOutput":{"hookEventName":"PreToolUse","additionalContext":"Check the v0 surface map in CLAUDE.md before writing Vue logic — never hand-roll a primitive @vuetify/v0 already provides."}}'
    fi
    ;;
esac
exit 0

3. Let the Type Checker Catch What Slips Through

Hallucinated v0 APIs fail to compile. Run vue-tsc --noEmit (or tsc --noEmit) as a verification step before accepting agent-written code — it converts “plausible but wrong” into a hard error the agent can fix itself.

How It Works

All three files are produced at build time:

• llms.txt and llms-full.txt are generated by generate-llms-full.ts↗︎ — the first extracts titles and descriptions organized by category, the second inlines the complete content of every markdown page.

• SKILL.md is authored in the repository at skills/vuetify0/SKILL.md and copied to the docs root by copy-markdown.ts↗︎, together with every reference it links under /references/.