ChatGPT Plugins
A ChatGPT plugin is the package people discover, install, and share in ChatGPT and Codex. A plugin can contain skills, an MCP server, or both, plus optional in-chat UI. It is the distribution wrapper; the app you actually use is the MCP server inside it.
The 2023 plugin and the 2026 plugin are not the same thing
This term has been reused, and that is the single biggest source of confusion around it. The 2023 plugin was a manifest-based integration with its own store and a per-chat enable toggle; that generation was retired in favor of apps and connectors. The word came back in July 2026, when OpenAI migrated the App directory to the Plugin directory — but it now names something structurally different. A 2026 plugin is a package, not a single integration. Any tutorial written before mid-2026 that says "plugin" means the retired generation, and its steps no longer apply.
What a plugin actually contains
OpenAI's developer documentation defines plugins as "the packages people discover, install, share, and publish in ChatGPT and Codex." Its user-facing documentation lists six kinds of component a plugin can bundle, not two: skills (reusable workflow instructions the model loads when relevant), connectors (the connection you authorise to a tool like Slack or Google Drive), MCP servers (the services behind connectors — they define tools, enforce auth, and return structured results), browser extensions, hooks (commands that run at configured lifecycle points, which OpenAI says to review and trust before enabling), and scheduled task templates. Most coverage mentions only the first and third, which makes plugins sound narrower than they are. Custom UI remains optional throughout.
Where plugins live
ChatGPT and Codex share one universal plugin directory, so a published plugin surfaces the same listing in either product. Individual capabilities can still be surface-specific — a plugin can include hooks that only run in Codex. Publishing goes through the plugin submission portal on the OpenAI Platform, where the required permission is still labeled "Apps Management" — one of several places where the older vocabulary is still visible in the product.
Plugin, app, skill, or connector: which word to use
Use plugin for the installable package. Use app for the MCP-backed integration that connects ChatGPT to a real third-party service and does something in a conversation — Canva, Notion, Spotify. Use skill for the packaged workflow instructions inside a plugin. Use connector for the connection you authorise to a service — and note that a connector and an MCP server are the same layer described to two audiences: OpenAI's user docs say connectors are backed by MCP servers, so they are not alternatives. OpenAI's own docs still use "app" alongside "plugin" — the Apps SDK kept its name — so both terms are current, at different layers.
What we rank
ChatGPTAppsRank ranks MCP-backed plugins with a real in-chat experience — the layer most people mean when they say "app" or "plugin." We do not score skills-only plugins: a package of instructions with no service behind it is a different product shape, and scoring it on setup, privacy clarity, or reliability would be measuring nothing. The distinction is drawn in full on the apps vs plugins page.