A2UI Atomic Catalog for UI

VerifiedCaution

Renders real interactive UI like charts, gauges, and dashboards via MCP instead of raw HTML.

Sby Skills Guide Bot
DevelopmentIntermediate
209/6/2026
Claude CodeCursorWindsurfCopilotCodex
#ui-rendering#mcp#data-visualization#a2ui#atomic-components

Recommended for

Our review

This skill lets an AI agent compose real user interfaces from a fixed catalog of 474 typed atomic components, rendered through an MCP server instead of hand-writing HTML.

Strengths

  • Typed, fixed component vocabulary eliminates hallucinated field names and validation errors.
  • Renders inline on MCP Apps-capable hosts or as a shareable link elsewhere.
  • No auth, no signup, free and MIT licensed, with strict JSON schemas per atom.
  • Loads only the needed catalog slice, keeping agent context lean.

Limitations

  • Does not fetch arbitrary live data; only sources explicitly declared by the catalog.
  • Requires a host that can display HTML or open a link.
  • The shared preview renderer is rate-limited.
When to use it

Use this skill when the user needs to see an interactive visual — chart, gauge, dashboard, decision tree — and the host can display rendered HTML or open a link.

When not to use it

Do not use it for plain prose answers, for data retrieval, or on hosts that cannot display HTML.

Security analysis

Caution
Quality score85/100

The skill uses network access to an external MCP server for legitimate UI rendering purposes. It does not instruct destructive, obfuscated, or explicitly exfiltrating actions, but the potential for data to be sent to a third-party renderer warrants caution.

Findings
  • Instructs connecting to an external MCP server at a2uicatalog.ai and sending data to its renderer via preview_url, which could expose user input if not explicitly approved or sanitized.
  • No allowed-tools are declared, but the skill guides network access to a third-party service without built-in user confirmation steps.

Examples

Interactive dashboard
List the available catalogs and render a small executive dashboard with a KPI chart and a gauge using the A2UI atomic catalog.
Decision tree visualization
Show me a decision tree for troubleshooting a login issue. Use get_catalog to check the decision-tree atom fields, then render_surface inline.
Stepper UI for onboarding
Create a 3-step onboarding stepper widget describing the setup process and generate a preview link.

name: a2ui-atomic-catalog description: Renders real interactive UI — charts, gauges, steppers, dashboards, decision trees — for AI agents instead of generating raw HTML. Use this skill when a user needs to SEE something rather than read a description of it, and the host (Claude, another MCP client, a Gemini Enterprise agent) can display rendered visuals or open a link. Do not use for plain prose answers, for data retrieval (this renders data, it does not fetch it beyond a few declared sources), or on hosts that cannot display HTML.

A2UI Atomic Catalog

474 typed UI atoms an AI agent composes into real rendered interfaces — web, Google Meet, Apps Script, Google Chat and MCP Apps — instead of writing HTML by hand. The agent names a component from a fixed vocabulary; a renderer that already knows that component draws it.

Connect

MCP server, no auth, no signup, no API key:

https://a2uicatalog.ai/mcp

POST speaks JSON-RPC 2.0 (MCP Streamable HTTP). GET with Accept: application/json returns a machine-readable descriptor listing every tool, current rate limits, and documentation links.

How to use it

  1. list_catalogs — pick the catalog slice the task actually needs. Do not load the whole 474-atom vocabulary into context for a one-chart request.
  2. get_catalog — read the real field contracts for that slice. Never guess a field name; a hallucinated field is a validation error, not a broken render.
  3. Render, by host capability:
    • render_surface — MCP Apps-capable hosts (Claude.ai and others): renders inline in the conversation, no URL.
    • preview_url — any other host: returns a shareable link (rate-limited demo renderer).
    • make_surface_url — render against the caller's own deployed renderer instead of the shared demo one; no rate limit.

When this is the wrong tool

  • The answer is better as prose than a visual.
  • The host cannot display HTML or open a link.
  • The task needs live data this catalog doesn't already declare a source for — this skill renders data, it does not fetch arbitrary data.

Reference

| Document | URL | |---|---| | OpenAPI specification | https://a2uicatalog.ai/openapi.json | | Full atom vocabulary | https://a2uicatalog.ai/spec.json | | Strict per-atom JSON Schema | https://a2uicatalog.ai/catalogue/atoms-json-schema.json | | Agent overview | https://a2uicatalog.ai/llms.txt | | A2A agent card | https://a2uicatalog.ai/.well-known/agent-card.json |

Free, MIT licensed, no signup. Independent, unofficial project — not affiliated with, endorsed by, or sponsored by Google or Anthropic. A2UI is Google's protocol; MCP is Anthropic's. Source: https://github.com/a2uicatalog/a2ui

Related skills