# What is A2UI: Google's open protocol for agents that answer with UI

> A2UI 1.0 shipped on November 11, 2026, with renderers for React, Angular, Lit, Flutter and SwiftUI and adopters including Atlassian.

URL: https://claudegpt.si/news/what-is-a2ui  
Published: 2026-11-15T02:52:52.698Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Tools  
Topics: Google, AI agents, Developer tools, Open source, MCP  
Publisher: ClaudeGPT Si

## Key takeaways

- A2UI is an open protocol from Google that lets an AI agent send a screen as JSON data, which the app then draws with its own components.
- Version 1.0, the first stable release, came out on November 11, 2026, with two-way function calls, accessibility support and agent SDKs for Python, TypeScript and Dart.
- Renderers exist for React, Angular, Lit, Flutter and SwiftUI, with Jetpack Compose at 0.9.1, and the project is open source under Apache 2.0.

A2UI (short for Agent-to-User Interface) is an open protocol from Google that lets an AI agent answer with an interactive screen instead of a block of text. The agent sends a description of the screen as data: a card here, a date picker there, a button. The app then draws it with its own native components. Version 1.0, the [first stable release](https://developers.googleblog.com/a2ui-10-build-delightful-fast-reliable-agentic-experiences/), came out on November 11, 2026.

That split is the whole idea. The agent decides what to show. The app decides how it looks and what is allowed. A booking agent can put a flight picker on screen in a banking app, a chat window and an iPhone app, and each one renders it in its own style.

## Why Google made A2UI

Google [introduced A2UI](https://developers.googleblog.com/introducing-a2ui-an-open-project-for-agent-driven-interfaces/) as an open project on December 15, 2025. Its reasoning was practical. Agents increasingly run on someone else's server, so they can't reach into an app and change its interface. And text is a clumsy way to do many tasks. Booking a table by typing back and forth with a chatbot takes longer than tapping a time slot.

The other option is to let the agent write HTML and JavaScript and run it in a sandbox. That works, but the result doesn't match the app's design, and the app is running code a model wrote. A2UI avoids both problems. In Google's words from the 2025 launch post, "A2UI is a declarative data format, not executable code."

## How A2UI works

A2UI messages are JSON. The screen is a flat list of components that point to each other by ID. That shape is easy for a language model to write a piece at a time, and easy to update as a conversation goes on. The UI can stream in while the agent is still working.

Three rules keep the app in charge:

- **A locked catalog.** The app registers the components an agent may use, such as Card, ChoiceChip, DatePicker and Button. The agent can only place and configure those. It can't invent new ones.
- **Native rendering.** Each app draws the components with its own widgets, so the result follows its design system.
- **Server-side binding.** Sensitive values can be bound on the server, so they never pass through the model.

The project's [GitHub README](https://github.com/a2ui-project/a2ui) sums it up as expressing UI "using data, not code." It is open source under the Apache 2.0 license, and the repository had about 17,400 stars when it was checked on November 15, 2026.

## What's new in A2UI 1.0

The [1.0 announcement](https://developers.googleblog.com/a2ui-10-build-delightful-fast-reliable-agentic-experiences/) on the Google Developers Blog lists these changes:

- **A stable protocol.** Earlier versions were 0.8 and 0.9. 1.0 is the first one Google calls stable, and it links an upgrade guide from 0.9.
- **Function calls in both directions.** A component can call back to the agent while the user works with it. Google's example is a reservation picker that checks availability on the spot.
- **Accessibility support** in the protocol itself.
- **Extensions** for extra metadata.
- **Agent SDKs** for Python, TypeScript and Dart that help generate the agent's instructions, translate messages and correct errors.
- **A2UI Macros**, in Python only, which expand components and bind data in code.
- **Faster formats.** A2UI Express is a compact format in which the model writes one line per component. Google also says "system one" decision models, which pick components instead of writing text, can drive A2UI. (The site has an [explainer on System One models](https://claudegpt.si/news/what-is-a-system-one-model).)

## Where A2UI runs

| Platform | Renderer | Status on November 15, 2026 |
| --- | --- | --- |
| Web | React, Angular, Lit | 1.0, on npm |
| Web, mobile, desktop | Flutter | 1.0 |
| iOS | SwiftUI | 1.0 |
| Android | Jetpack Compose | 0.9.1, 1.0 "coming soon" |
| ByteDance Lynx | Community renderer | Supports 0.9, moving to 1.0 |

Google says the web and Flutter renderers also work inside MCP Apps, so an A2UI screen can appear in assistants that support MCP Apps, such as Claude and Gemini.

## Who uses A2UI

Google names several adopters in the 1.0 post. Atlassian is bringing A2UI to Jira, where agents can ask teams for follow-ups through interactive cards. Alation builds data feeds of metric cards and charts that render both in Claude and in its own product. ByteDance released the Lynx renderer. Bloomberg is named as building with it, without details.

Inside Google, Gemini Notebook uses A2UI for its slide deck and video overview cards. Organizations can register A2UI agents in Gemini, which supports version 0.9.1, and Gemini Enterprise is being upgraded to 1.0.

## A2UI vs MCP Apps and AG-UI

These names come up together, and they do different jobs.

- **MCP Apps** deliver a finished HTML page that runs in a sandboxed frame. A2UI sends a blueprint that the host app draws itself, so it inherits the app's look. Google now treats the two as complementary: A2UI screens can run inside MCP Apps, and the README lists embedding MCP Apps inside A2UI as roadmap work.
- **AG-UI** is CopilotKit's protocol for connecting an agent to a front-end app. Google's 2025 post called it complementary: AG-UI carries the conversation, and it supports A2UI for drawing the agent's responses. CopilotKit also runs the A2UI Playground.
- **A2A**, Google's agent-to-agent protocol, can carry A2UI messages to a client. The project ships an `a2ui-a2a` Python package for that.

## How to try A2UI

The [A2UI site](https://a2ui.org/) has a quickstart it says runs in five minutes, and the docs for each renderer. The A2UI Playground lets you type a request and watch a model write the UI. For a real app, the order is: pick the renderer for your platform, register the components agents may use, then add the agent SDK on the server.

The thing to watch next is the Android renderer reaching 1.0, and whether other assistants beyond Google's own adopt it as their way of drawing agent answers.

## Frequently asked questions

### Is A2UI open source?

Yes. A2UI is published under the Apache 2.0 license, and its code, renderers and SDKs are on GitHub in the a2ui-project/a2ui repository.

### What is the difference between A2UI and MCP Apps?

MCP Apps deliver finished HTML that runs in a sandboxed frame. A2UI sends a description of the screen that the host app draws with its own native components. Google says A2UI screens can also run inside MCP Apps.

### What is the difference between A2UI and AG-UI?

AG-UI, from CopilotKit, connects an agent to a front-end app. A2UI describes the interface the agent wants to show. Google calls them complementary, and AG-UI supports A2UI for rendering agent responses.

### Does A2UI work on Android and iOS?

Yes. iOS has a SwiftUI renderer and Flutter covers both platforms. The Jetpack Compose renderer for Android was at version 0.9.1 on November 15, 2026, with 1.0 announced as coming soon.

## Sources

- [A2UI 1.0: Build delightful, fast, reliable, agentic experiences](https://developers.googleblog.com/a2ui-10-build-delightful-fast-reliable-agentic-experiences/), Google Developers Blog
- [Introducing A2UI: An open project for agent-driven interfaces](https://developers.googleblog.com/introducing-a2ui-an-open-project-for-agent-driven-interfaces/), Google Developers Blog
- [A2UI](https://a2ui.org/), A2UI project
- [a2ui-project/a2ui](https://github.com/a2ui-project/a2ui), GitHub
