A full-stack TypeScript framework for AI-embedded views, enabling cross-platform development for ChatGPT Apps and MCP Apps.
Skybridge is a full-stack TypeScript framework developed by Alpic, designed specifically for embedding rich interactive views within AI conversations. Its core advantage lies in achieving end-to-end type safety between server-side tool definitions and client-side React views (using tRPC-style inference), and enabling bidirectional synchronization of UI state and LLM Context via the data-llm attribute. The framework encapsulates a protocol abstraction layer at its core, allowing a single codebase to run simultaneously on ChatGPT Apps and MCP-compatible clients (like Goose, VSCode).
Core Capabilities#
- tRPC-style End-to-End Type Inference: Tool Schemas defined on the Server automatically flow to Client Hooks, with full-chain autocomplete for tool inputs and outputs.
- React Query-style Hooks: Provides APIs like
useToolInfo, supportingisPending,isError, and callbacks for familiar async state management. - View → Model Sync: Synchronizes frontend UI state to the LLM via custom
data-llmHTML attributes, achieving "single source of truth across two surfaces". - View → Tool Calls: Supports triggering server-side tool calls directly from UI components.
- Dual Platform Compatibility: A single codebase supports both ChatGPT Apps SDK and MCP-compatible clients.
- Vite HMR & Local Simulation: Built-in Vite plugin with hot reload, Debug Traces, and local DevTools simulator—no repeated reinstalls needed.
Architecture#
Adopts a pnpm workspace-managed full-stack monorepo architecture:
skybridge/server: Server-side core built on top of the MCP SDK, providingregisterView/ tool definition APIs with built-in Zod schema validation and end-to-end type inference.skybridge/web: Client-side core—a collection of React hooks that consume server-derived types, with underlying adapters for both ChatGPT Apps SDK and MCP Apps protocols.- Dev Environment: Vite plugin system providing HMR, DevTools simulator, and optimized builds.
Key design patterns include a protocol abstraction layer (encapsulating unified view/tool registration APIs above the MCP protocol while isolating ChatGPT Apps SDK adaptation) and a type flow mechanism (Server Schema declarations penetrate directly to Client component props via the TypeScript type system).
Quick Start#
npm create skybridge@latest
Or add to an existing project:
npm i skybridge
pnpm add skybridge
yarn add skybridge
bun add skybridge
deno add skybridge
Engineering Practices#
Uses Biome for unified lint/format; CI/CD via GitHub Actions; automated dependency updates through Renovate; repository includes AGENTS.md and CLAUDE.md for native support of AI coding agent workflows like Claude Code.
Examples & Ecosystem#
Includes over a dozen production-grade examples: flight booking, e-commerce carousels, map exploration, auth integration (Auth0/Clerk/Stytch/WorkOS), generative UI, and more. Known production users include Awaze (cottages.com).