google-labs-code/stitch-sdk

What it solves

It allows developers to programmatically generate, edit, and iterate on UI screens using text prompts. It bridges the gap between a text-based design idea and actual code by providing the resulting HTML and screenshots of the generated screens.

How it works

The SDK interacts with a backend service to create projects and generate screens based on natural language descriptions. It supports iterative design through editing prompts and the creation of multiple design variants (adjusting layout, color schemes, etc.). It also provides a "Tool Client" and integrations for the Vercel AI SDK and Google Agent Development Kit (ADK), allowing AI agents to autonomously call design tools to build interfaces.

Who it’s for

• Frontend Developers who want to rapidly prototype UI screens from text.
• AI Agent Developers building autonomous designers or orchestration scripts that can create and modify user interfaces.
• Product Designers looking to programmatically explore different visual variants of a screen.

Highlights

• Text-to-UI Generation: Create full screens from prompts and retrieve HTML and image assets.
• Iterative Editing: Modify existing screens using follow-up text prompts.
• Design Variants: Generate multiple versions of a screen with specific controls over creative range and aspects like fonts or colors.
• Agent-Ready: Built-in support for MCP (Model Context Protocol) tools, making it easy to integrate with LLM frameworks like Vercel AI SDK and Google ADK.
• Design System Integration: Ability to create and apply consistent visual themes across projects.