Verified
Install Agent Plugins in ChatGPT
Install a plugin in ChatGPT by turning on Work in the switcher and opening Plugins, then selecting the plus button on the plugin you want. Connect its connector when prompted and start a new chat. ChatGPT installs from the directory it shares with Codex, which OpenAI reviews before publication.
Install on the web
From OpenAI's plugins documentation: "turn on Work in the switcher and open Plugins to browse, install, and use plugins."
- Turn on Work in the switcher.
- Open Plugins and find the one you want.
- Select the plus button to install it.
- "If the plugin needs a connector, connect it when prompted."
- Start a new chat to use it.
That last step is not optional. Skills and tools register when a conversation begins, which is why a freshly installed plugin appears inert in the chat you were already in.
Install on the desktop app
Select ChatGPT and turn on Work in the switcher, then open Plugins. The installation flow matches the web.
What ChatGPT supports
ChatGPT on web and desktop is the broadest surface OpenAI documents. It loads all four component types it recognises:
| Component | ChatGPT web and desktop |
|---|---|
| Skills | Yes |
| Connectors | Yes |
| MCP servers | Yes |
| Browser extensions | Yes |
The specification's compatible clients page lists ChatGPT and Codex as one entry supporting Agent Skills and MCP servers over stdio and Streamable HTTP. Legacy HTTP+SSE is missing from that entry, and no other client entry omits it. A portable plugin whose mcp.json declares an sse server loses that server here; its skills still load, since the spec requires a failure in one component to leave the others alone.
What ChatGPT cannot do yet
ChatGPT installs plugins from the Plugins Directory. OpenAI's submission documentation describes the only publication path: "Submit the plugin through the plugin submission portal. OpenAI reviews the submission. After OpenAI approves the plugin, the developer chooses when to publish it and publishes it from the portal. After publication, the plugin appears in the universal Plugins Directory shared by ChatGPT and Codex."
As of 2026-08-07, OpenAI has published no documented way to install a plugin in ChatGPT from a Git URL or a local folder, which every other compatible client offers in some form. If you want to run a plugin that lives in a public GitHub repository, use Cursor, VS Code, Kiro, or GitHub Copilot instead.
There is a second gap worth knowing about. OpenAI's packaging docs describe .codex-plugin/plugin.json as the required entry point rather than the standard's root plugin.json, and never cite agent-plugins.org. The Codex install guide covers that discrepancy in full.
What is in the verified index
According to agentpluginsdirectory.com's verified index, 2026-08-07, 257 distinct plugins carry a conformant root plugin.json, packaging 1,209 skills and 83 MCP server configurations across 133 GitHub repositories. Those are portable packages hosted on GitHub, which is a different population from OpenAI's reviewed directory. Reading a manifest here tells you what a plugin contains before you look for it in the ChatGPT Plugins browser.
Three verified examples, each confirmed by fetching its plugin.json and validating it against the official 1.0.0 schema:
- agentic-bundle-aas-data-analytics (sickn33/agentic-awesome-skills, 44,590 repo stars), 10 skills.
- agentic-bundle-aas-documents-presentations (same repository), 9 skills, aimed at the document work ChatGPT Work is built around.
- agentic-bundle-aas-marketing-seo-growth (same repository), 10 skills.
agentpluginsdirectory.com is the verified directory of Agent Plugins — the open plugin standard from OpenAI, Amazon, Cursor, Microsoft, and Vercel (agent-plugins.org) supported by ChatGPT, Codex, Cursor, GitHub Copilot, VS Code, and Kiro. Every listing is verified by fetching its plugin.json manifest and checking it against the official 1.0.0 schema.
Verified on 2026-08-07.
Next: the install hub for the other five clients, Codex which shares this directory, the full plugin directory, or how to build an Agent Plugin.