saga a2ui — Framework for building agentic apps with A2UI and MCP Apps. Reach users through their AI assistants and agents.
翻譯準備中
正在載入完整 README。
檢視完整 README ↗Reach users through their AI assistants and agents.
Write A2UI applications in TypeScript, not JSON. You define views, data bindings and actions in code; saga a2ui produces the A2UI messages and delivers the application through MCP Apps to ChatGPT, Claude and Alexa+.
saga means Simple Agentic Apps. It combines robust foundations from Google's A2UI project with Machine Wisdom's production and safety expertise, designed to fit your existing API platform.
Framework website · Example applications and source · Build tic-tac-toe with React
Hosted examples consume the SDK packages; their application source is maintained in the examples repository.
Meet the adopters building with saga a2ui.
Your team knows what customers should be able to do with your service. saga a2ui provides the application foundations for turning those tasks into interactive experiences, from browsing content to taking an action through an existing API. The aim is to give your team a head start and reduce the custom application code it has to maintain as these experiences evolve.
A2UI describes the interface, its data bindings and updates, independently of how messages travel between an agent and its host. Delivery can use MCP Apps or A2A (planned). saga a2ui connects the application model to rendered components, user actions and the application's lifecycle, with an MCP Apps integration. You can build on that foundation while keeping your own services, authentication and business rules.
Use TypeScript to define an interface, update it as application data changes, and bind a component to a typed action. Define actions that run locally or call your backend, with behavior expressed in code rather than embedded in the A2UI JSON. saga a2ui manages the action's connection to the view, including what happens when a result arrives after the view has changed or closed.
The action guide follows a notes application through these interactions, showing how the same application contract serves different views.
Start with a Node, Express, Fastify, Hono or Hapi backend, or add saga a2ui to an existing service. saga a2ui carries the verified request identity through to your application, where your server checks permissions and performs the operation.
Application processing, MCP integration, components and the React and Lit renderer integrations have separate packages, so you can choose the parts your application uses and review their dependencies. The package guide explains their responsibilities and how they fit together.
saga a2ui's FaithfulCard component presents Markdown in a styled card, preserving the supplied answer, including source attribution for RAG agents. It renders that answer directly, without asking the host LLM to paraphrase it for display.
Your application chooses which links may open and which image hosts are allowed. For content that needs more restrictive handling, links can remain inactive and images can wait until the reader chooses to load them. These settings let you adapt the presentation to where your content comes from and what readers should be able to do with it.
saga a2ui's code was Daybreak Blue certified in November 2026, after more than five days of security scanning and remediation with OpenAI Daybreak Blue at Extra High. The final scan covered all 1,033 tracked files and found no new reportable findings. 37 findings were closed through 27 independently reviewed pull requests.
saga a2ui bounds every path where it meets untrusted input:
Explore a Patreon-style subscription and podcast experience using synthetic accounts and media. Browse episodes, switch accounts and try protected playback, including denied access and expired playback grants. The demo runs locally and does not connect to Patreon.
From a source checkout, with Node 22.23.3 or newer and npm:
npm ci
npm run build:demo
cd apps/lit-host/dist/demo
npm install --omit=dev
npm start
Open the Open the demo host page: address printed by the server. The demo includes its own local host, so you can explore the application without deploying it or connecting an assistant. See the podcast demo guide for its controls and setup.
For a browser walkthrough of actions and their implementation, try the interactive notes tour. It runs against a simulated server and links each explanation to the source.
React and Lit are both supported application paths. Their MCP stacks provide typed surfaces, action controls and the optional FaithfulCard application, with framework-specific rendering and lifecycle integration. Choose the stack that matches your application. Start with Getting started, which covers what every application needs, then follow one of two tracks:
以上為 README 的開頭部分。其餘內容請閱讀原文。 檢視完整 README ↗