A2UI Atomic Catalog

VerifiedCaution

Renders interactive UI components (charts, gauges, dashboards) via MCP instead of raw HTML. Use when users need visuals on capable hosts.

Sby Skills Guide Bot
DevelopmentIntermediate
108/28/2026
Claude CodeCursorWindsurfCopilotCodex
#a2ui#ui-atoms#mcp#visualization#interactive-ui

Recommended for

Our review

This skill uses an MCP server to let an agent compose real interactive interfaces from a catalog of 474 predefined UI atoms, instead of hand-writing HTML.

Strengths

  • Renders genuinely interactive UI such as charts, gauges, steppers, dashboards, and decision trees
  • Fixed atom vocabulary reduces invalid field names and hallucinated markup
  • Multiple rendering paths: inline in MCP Apps, shareable preview URL, or the caller's own renderer
  • Free, no auth, no signup, MIT licensed

Limitations

  • Only renders data — it does not fetch arbitrary live data beyond documented sources
  • Requires a host that can display HTML or open a link
  • The shared preview_url renderer is rate-limited
When to use it

When the user needs to see a visual representation (chart, dashboard, decision tree) instead of reading a text description.

When not to use it

When a prose answer is clearer, the host cannot display rendered visuals, or the task depends on live data outside the catalog's declared sources.

Security analysis

Caution
Quality score85/100

The skill itself does not contain destructive commands or obfuscated payloads, but it relies on a remote MCP server for rendering. This involves network calls to an untrusted third-party endpoint and potential transmission of user data, so caution is warranted.

Findings
  • Skill instructs the agent to connect to an external MCP server (a2uicatalog.ai) over the network, which may involve sending UI component data to a third-party service.
  • No authentication or signing is used, so the service could be spoofed or logs could be captured by the host operator; skill does not warn against sending sensitive data.

Examples

Interactive gauge dashboard
Use the a2ui atomic catalog to render a small dashboard with a gauge showing CPU usage at 67%, a stepper to set a warning threshold, and a line chart of the last 24 hours.
Login troubleshooting decision tree
Render a decision tree for troubleshooting a login failure: check password, then 2FA, then account lockout, with yes/no branches and a final recommendation.
Project progress stepper
Create an interactive stepper with four steps for onboarding: account creation, profile setup, billing, and launch, and mark the current step as 'profile setup'.

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