context7.json

{
  "$schema": "https://context7.com/schema/context7.json",
  "url": "https://context7.com/alpic-ai/skybridge",
  "public_key": "pk_QmCV6r1eoYWNTIq1jxP0h",
  "projectTitle": "Skybridge",
  "description": "Fullstack TypeScript framework for building ChatGPT and MCP Apps with type safety, React hooks, and HMR",
  "folders": ["docs/docs", "examples"],
  "excludeFolders": [
    "node_modules",
    "dist",
    "src",
    "**/__snapshots__",
    "docs/node_modules",
    "docs/.docusaurus",
    "packages",
    "examples/**/.env",
    "examples/**/node_modules"
  ],
  "excludeFiles": [
    "docusaurus.config.ts",
    "sidebars.ts",
    "vercel.json",
    "package.json",
    "pnpm-lock.yaml",
    "tsconfig.json",
    "README.md",
    "CONTRIBUTING.md",
    "LICENSE",
    "biome.json",
    "vitest.config.ts",
    "pnpm-workspace.yaml"
  ],
  "rules": [
    "Use TypeScript for full type safety with Skybridge",
    "Use method chaining when defining your MCP server (e.g., new McpServer().registerWidget().registerTool()) to enable type inference with generateHelpers",
    "Export your server type as 'AppType' (export type AppType = typeof server) to use with generateHelpers for end-to-end type inference",
    "Use React hooks from skybridge/web (useToolInfo, useCallTool, useWidgetState, useLayout, useUser) instead of raw window.openai APIs for better state management",
    "Use useToolInfo for initial data hydration - it provides the data from the tool call that rendered the widget",
    "Use useCallTool for user-triggered actions only - do NOT call tools on mount, pass initial data through structuredContent instead",
    "Use the data-llm attribute on DOM elements to sync UI state with the model, enabling contextual responses",
    "Widget endpoint names must match widget file names exactly (e.g., 'pokemon-card' endpoint requires 'pokemon-card.tsx' file in web/src/widgets/)",
    "Use Zod schemas for input and output validation in widget registration (inputSchema and outputSchema)",
    "Use generateHelpers to create fully typed hooks with autocomplete - import AppType from server and call generateHelpers<AppType>()",
    "Prefer skybridge/server's registerWidget over raw MCP tool registration for better type inference and widget management",
    "Use the local dev environment with HMR (pnpm dev) for faster development iterations without reinstalling the app in ChatGPT",
    "Widgets run in an iframe with CSP restrictions - useCallTool is the preferred way to fetch additional data as it's guaranteed to be allowed",
    "Use React Query-inspired patterns: check isPending, isSuccess, isError states from hooks, and use callbacks (onSuccess, onError) for side effects"
  ]
}