Codex plugins, visually explained

Codex v0.117.0 (released yesterday) now supports plugins.

OpenAI just added plugin support to Codex, and I made this because I was reading the official docs trying to get the mental model straight in my own head.

Plugin overview

A plugin is a packaged Codex setup: skills, apps / connectors, and MCP servers bundled into one reusable unit.

My rough read:

Skills = reusable workflows
Apps / connectors = service integrations
MCP servers = external tools + context
Plugins = the installable bundle that packages those together

What can go inside a plugin

What can live inside the bundle: skills for know-how, apps / connectors for integrations, and MCP servers for tools or external context.

Minimal plugin structure

The simplest plugin shape: one manifest plus whatever optional pieces the workflow actually needs.

I'll admit I'm still a bit confused about the boundary between apps and MCPs, since they seem to overlap.

My current read is that MCP is the more general connection layer, while apps feel more like service integrations that show up more directly in the product. For example, all the services I've connected in ChatGPT automatically appear in Codex.

Still, this feels like a useful change. It makes Codex workflows much easier to package and share across a repo or a personal setup.

Installing from slash plugins

The user-facing entry point seems to be /plugins, where you can browse and install available plugin bundles.

Plugin details

Once you open one, you get the plugin details view and can inspect what the bundle contains.

Plugin and skill insertion

Installed plugins then show up in the normal flow of using Codex, including insertion and discovery moments.

Repo vs personal install scope

One distinction that matters: some installs are repo-scoped and some are personal, depending on where you want the workflow to live.

If someone has a cleaner mental model for apps vs MCPs, I'd love to hear it.

Codex plugins, visually explained

Docs

Subscribe for Updates