chatgpt-apps

Scaffold and validate ChatGPT Apps SDK applications combining MCP servers and widget UIs.

Updated Jul 3, 2026
One-click install
npx skills add https://github.com/CHENHUI-X/toolbox --skill chatgpt-apps-chenhui-x
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: chatgpt-apps
Source: https://github.com/CHENHUI-X/toolbox/tree/main/codex-skill/codex/chatgpt-apps
Command: npx skills add https://github.com/CHENHUI-X/toolbox --skill chatgpt-apps-chenhui-x

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires @modelcontextprotocol/ext-apps, @modelcontextprotocol/sdk, zod, tsx, typescript, and includes scripts (resource) and references (resource) components.

What problem does it solve? Building ChatGPT Apps SDK applications requires coordinating an MCP server, widget UI, tool metadata, CSP settings, and deployment steps, and doing this from scratch often produces repos that drift from current OpenAI documentation. ## Core Features & Use Cases - Docs-First Scaffolding: Fetches current Apps SDK docs via the openai-docs skill before generating code, then classifies the request into one of five app archetypes (tool-only, vanilla-widget, react-widget, interactive-decoupled, submission-ready). - Server and Widget Generation: Produces MCP servers with registered tools, annotations, and UI resources, plus widgets using the MCP Apps bridge with optional window.openai compatibility APIs. - Validation and Deployment Guidance: Enforces a minimum working repo contract, provides a validation ladder, and covers local tunnel testing, ChatGPT Developer Mode setup, hosting, and public submission checklists. - Use Case: Ask for a ChatGPT app that searches a company knowledge base; the skill classifies it as tool-only, implements the standard search and fetch tools, scaffolds the MCP server, and reports exactly which validation checks were run. ## Quick Start Use $chatgpt-apps with $openai-docs to scaffold a ChatGPT app for my use case with a TypeScript MCP server and a vanilla widget.

Frequently Asked Questions about chatgpt-apps

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I build a ChatGPT app with the Apps SDK?

Start by classifying the app archetype, then fetch current Apps SDK docs before writing code. Plan tools with clear schemas and annotations, scaffold the MCP server with a registered widget resource, and build the widget on the MCP Apps bridge with optional window.openai extensions.

How do I test a ChatGPT app locally during development?

Run the MCP server on http://localhost:<port>/mcp and expose it with a public HTTPS tunnel such as ngrok. In ChatGPT, enable Developer Mode under Settings, create a new app with the tunneled URL plus /mcp path, and refresh the app after tool or metadata changes.

Should my ChatGPT app use the MCP Apps bridge or window.openai?

Use the MCP Apps bridge first for portability: ui/notifications/tool-result, tools/call, and ui/message cover baseline behavior. Add window.openai APIs only for ChatGPT-specific features like file upload, modals, display mode changes, or theme signals.

When should a ChatGPT app implement the standard search and fetch tools?

Implement the standard search and fetch tools when the app is connector-like, data-only, sync-oriented, or meant for company knowledge and deep research. These read-only tools follow an exact schema so ChatGPT can cite results with canonical URLs.

What is required to submit a ChatGPT app to the public directory?

Submission requires a public production HTTPS endpoint, accurate CSP metadata, org verification with Owner role, and artifacts like app metadata, screenshots, privacy policy, and test prompts. Private or internal apps should stay in Developer Mode instead of submitting.