Upstream Example Workflow

This keeps the skill aligned with current docs and maintained example code while still preserving a low-dependency fallback when examples are not a good fit.

Choose The Right Source

1. Official OpenAI examples

Prefer these when:

• The app is clearly ChatGPT-facing
• The user wants a polished UI or React component
• The task involves file upload, modal flows, display-mode changes, or other ChatGPT extensions
• The docs/examples page already shows a similar interaction pattern

Typical sources:

• https://developers.openai.com/apps-sdk/build/examples/
• https://github.com/openai/openai-apps-sdk-examples
• https://developers.openai.com/apps-sdk/quickstart/ for the smallest vanilla baseline

2. @modelcontextprotocol/ext-apps examples

Prefer these when:

• The user needs a lower-level MCP Apps baseline
• Portability across MCP Apps-compatible hosts matters more than ChatGPT-specific polish
• You want version-matched examples close to the installed @modelcontextprotocol/ext-apps package shape

This follows the same basic idea as the upstream create-mcp-app skill: use maintained examples as the starting point, then adapt them.

Typical examples from upstream flows:

• examples/demo-vanilla-html
• examples/demo-react-simple
• examples/demo-connectors-api

3. Local fallback scaffold

Use scripts/scaffold_node_ext_apps.mjs when:

• No close upstream example exists
• The user wants a tiny Node + vanilla HTML starter
• Network/example retrieval is undesirable
• You need a throwaway starter to patch quickly during a live coding task

Do not prefer the local scaffold just because it is available. It is the fallback, not the default.

Adaptation Rules

• Copy the smallest matching example, not the entire showcase app.
• Remove unrelated demo tools, assets, and routes immediately.
• Keep the upstream file structure when it is already clean and docs-aligned.
• Reconcile the copied example with the current docs before finishing:
  • tool names and descriptions
  • annotations (readOnlyHint, destructiveHint, openWorldHint, idempotentHint when true)
  • _meta.ui.resourceUri and optional _meta["openai/outputTemplate"]
  • resource _meta.ui.csp, _meta.ui.domain, and openai/widgetDescription
  • URI versioning for template changes
  • local run/test instructions
• State which example you chose and why.
• If you rely on upstream code, note the source repo and branch/tag/commit when practical; avoid silently depending on a floating example shape for long-lived work.

Minimal Selection Heuristic

• If the user asks for React + polished UI, start with official OpenAI examples.
• If the user asks for vanilla HTML + tiny demo, start with the quickstart example; use the local fallback scaffold only if the quickstart is still too opinionated or unavailable.
• If the user asks for portable MCP Apps wiring, start with @modelcontextprotocol/ext-apps examples.
• If the user already has an app, adapt their code directly instead of importing a new example.