AgenticFrameworksMCP

A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs

by Google (with CopilotKit and the A2UI open-source community)

IntermediateDocumentationFree~5 min quickstart; roughly 3-4 hours self-paced to work through the concepts and guides (estimate)

Learn how AI agents can return rich, native UIs as safe declarative JSON instead of text or executable code.

Start LearningAdded Sep 26, 2026 · Updated Sep 26, 2026

Overview

A2UI (Agent-to-User Interface) is an Apache-2.0 protocol created by Google, with contributions from CopilotKit and the open-source community, that lets an agent describe a user interface as declarative JSON which the client renders with its own native components, so no agent-generated code is ever executed. The documentation at a2ui.org is organised into an Introduction and FAQ, a Quickstart, the A2UI Composer, Concepts, Guides, Reference, versioned Specifications (v0.8 through the v1.0 release candidate; v0.9.1 is the current production release), Ecosystem and a Roadmap. The Concepts section covers the glossary, data flow, components and structure, data binding, catalogs, transports and actions: v0.9 uses four message types (createSurface, updateComponents, updateDataModel, deleteSurface), represents component trees as an adjacency list suited to incremental streaming updates, binds components to state with JSON Pointer paths, and restricts agents to a catalog of pre-approved components. The Guides cover client setup, agent development, using A2UI with any agent framework or harness, renderer development, defining your own catalog, authoring custom components, theming, and three A2UI + MCP guides. The quickstart runs a restaurant-finder demo in which a Python A2A agent calls Gemini and streams A2UI payloads to a Lit or Flutter client. The agent-development guide uses Google ADK and the A2uiSchemaManager from the a2ui agent SDK to generate a system prompt containing the schema and examples, then validates model output against the JSON schema before streaming. Transports include A2A, AG-UI, SSE and WebSockets.

At a Glance

Topic: Agentic
Level: Intermediate
Format: Documentation
Cost: Free
Duration: ~5 min quickstart; roughly 3-4 hours self-paced to work through the concepts and guides (estimate)
Provider: Google (with CopilotKit and the A2UI open-source community)
Hands-on: Yes — code/exercises
Certificate: None

What You’ll Learn

✓How the v0.9 message types createSurface, updateComponents, updateDataModel and deleteSurface build and update a UI surface
✓Why A2UI uses an adjacency-list component model and how that enables incremental streaming updates
✓How to bind components to application state with JSON Pointer paths for reactive, dynamic lists
✓How to define your own component catalog so agents can only request pre-approved widgets
✓How to prompt an LLM with A2uiSchemaManager and validate its A2UI JSON against the schema before rendering
✓How to carry A2UI over A2A, AG-UI, SSE or WebSockets and combine it with MCP servers
✓How to run the restaurant-finder quickstart with a Python agent, Gemini and a Lit or Flutter renderer
✓How to write a custom renderer or custom components and apply theming to agent-generated surfaces

Highlights

•Security model is the point: agents send data describing UI intent, never executable code, which suits cross-trust-boundary multi-agent setups
•Heavily adopted open project: the a2ui-project/a2ui GitHub repo shows about 16.5k stars and 1.3k forks (checked 2026-09-26)
•Framework-agnostic: renderers for Lit, Angular and Flutter (via GenUI SDK), with React, SwiftUI and Jetpack Compose on the roadmap
•Independent practitioners have published working ADK + A2UI v0.9 builds with schema validation, e.g. Mete Atamel's March 2026 walkthrough
•Versioned specs from v0.8 to a v1.0 release candidate are published side by side, so you can see what changed (v1.0 adds actionResponse)

Who It’s For

Best For

✓Agent developers who need to return forms, cards and lists rather than plain chat text
✓Frontend and mobile engineers building renderers for agent-generated interfaces
✓Teams already using A2A, AG-UI, Google ADK or MCP who want a standard UI payload format
✓Platform and security engineers evaluating how to show untrusted agent output safely

Prerequisites

•Working knowledge of LLM agents, tool calling and structured JSON output
•Python with uv for the sample agent, and Node.js 18+ (Lit client) or the Flutter SDK
•A Gemini API key from Google AI Studio to run the quickstart as written

FAQ

What is A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs?

A2UI's official documentation is the reference and tutorial set for Google's open Agent-to-User Interface protocol, aimed at developers building agents that must show forms, lists and cards to users. After working through it you can make an agent emit streaming A2UI JSON, validate it, and render it natively on web or Flutter clients.

Is A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs free?

A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs is free to access.

What level is A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs for?

A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs is aimed at a intermediate audience. Recommended background: Working knowledge of LLM agents, tool calling and structured JSON output, Python with uv for the sample agent, and Node.js 18+ (Lit client) or the Flutter SDK, A Gemini API key from Google AI Studio to run the quickstart as written.

How long does A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs take?

Expect roughly ~5 min quickstart; roughly 3-4 hours self-paced to work through the concepts and guides (estimate). Most learners work through it at their own pace.

What will I learn from A2UI Documentation — Agent-to-User Interface Protocol for Agent-Driven UIs?

You'll learn: How the v0.9 message types createSurface, updateComponents, updateDataModel and deleteSurface build and update a UI surface; Why A2UI uses an adjacency-list component model and how that enables incremental streaming updates; How to bind components to application state with JSON Pointer paths for reactive, dynamic lists; How to define your own component catalog so agents can only request pre-approved widgets; How to prompt an LLM with A2uiSchemaManager and validate its A2UI JSON against the schema before rendering; How to carry A2UI over A2A, AG-UI, SSE or WebSockets and combine it with MCP servers; How to run the restaurant-finder quickstart with a Python agent, Gemini and a Lit or Flutter renderer; How to write a custom renderer or custom components and apply theming to agent-generated surfaces.

Topics

a2uigenerative uiagent protocolsgoogle adka2aag-ui

Sources

This page was written from 6 sources, 2 on domains other than a2ui.org.