Module 11 · advanced · 1.5 hours

Plugins

Plugins are the distribution layer for reusable Codex and ChatGPT extensions. A plugin can bundle skills, an MCP server, or both. An MCP server can also return optional UI.

Plugin architecture

A plugin manifest describes identity and packaged capabilities. Keep the package cohesive: every included skill, connector, and asset should support one product story.

.codex-plugin/
├── plugin.json
├── skills/
├── .mcp.json
└── assets/

Skills, tools, and UI

Skills define reusable workflows. MCP servers provide typed tools and resources. An MCP server can also return optional UI for supported interactions.

Skill: Workflow knowledge
MCP: Tools and data
UI: Optional interaction surface

Test before distribution

Validate installation, metadata, trigger behavior, authentication, missing dependencies, and uninstall. Test on every host the plugin claims to support.

1. Validate package
2. Install locally
3. Test positive and negative triggers
4. Verify connector auth
5. Uninstall cleanly

Publish responsibly

Use accurate descriptions, minimal permissions, stable tool schemas, clear privacy behavior, and versioned release notes. Prefer a plugin when others should install the workflow; keep repo-only skills local.

Continue learning

Use the official OpenAI Codex documentation as the final reference for current commands, settings, and availability.

Return to the full curriculum.