ChatGPT Plugin Creator: from idea to a submitted plugin

This research library uses AI-assisted source research and drafting. Linked sources support product claims; analysis and proposed exercises are our interpretation. Unless an article documents a test and its results, do not read it as a hands-on review or an independently verified benchmark.

ChatGPT Plugin Creator is a built-in skill that scaffolds a plugin for ChatGPT and Codex from your skills or MCP server. You call it as @plugin-creator in ChatGPT Work mode or $plugin-creator in Codex. OpenAI introduced it at DevDay on 29 September 2026, together with a redesigned submission flow with clearer feedback and better ranking in the directory. All plans have access, and the plugins documentation is the primary source for every step below.

This guide takes one plugin from idea to a submitted listing. It follows the order the docs use: choose a shape, scaffold, test locally, then submit.

Choose the plugin shape first

A plugin can contain skills, an MCP server, or both. ChatGPT and Codex share one universal plugin directory, so a public plugin is published once and appears in both products. OpenAI's architecture guide lists four shapes:

Pick the smallest shape that works. One constraint shapes this choice: adding an MCP server to an existing skills-only plugin is not currently supported. If you expect to need tools later, include the server in your first upload.

Scaffold with ChatGPT Plugin Creator

For a plugin with an MCP server, connect the server first. In ChatGPT, open Settings, then Security and login, and turn on Developer mode. Go to ChatGPT Plugins, select the plus button, and enter your server URL. After ChatGPT creates the connection, copy the technical ID from the browser URL. It starts with plugin_asdk_app.

Give that ID to Plugin Creator. The docs show this prompt for Work mode:

@plugin-creator create a plugin for ChatGPT and Codex using my MCP server.
Use plugin_asdk_app_6a4c0062f3b88191855c0a80eac5d53d and name it Acme Support.
Include a personal marketplace entry so I can test it locally.

In Codex, use $plugin-creator with the same request. Plugin Creator writes the Codex compatibility layout. Only .codex-plugin/plugin.json is always created. It can add .mcp.json, .app.json, skills/, hooks/, scripts/ and assets/ when you ask.

Check two things after it runs. The registered MCP server mapping in .app.json should point at the right plugin_asdk_app ID. The apps field in plugin.json should point to ./.app.json. Add your skill folders under skills/.

Test the plugin locally in Codex

A marketplace is a JSON catalog of plugins. Plugin Creator can generate one, and you can keep adding entries to build a curated list for a repo or team. The ChatGPT desktop app, including Codex, reads a repo marketplace from .agents/plugins/marketplace.json and a personal one from ~/.agents/plugins/marketplace.json.

To track a marketplace from the Codex CLI instead of editing config.toml by hand:

codex plugin marketplace add ./local-marketplace-root
codex plugin marketplace list

Restart the ChatGPT desktop app, open the Plugins Directory, choose your marketplace, and install the plugin. Then build an evaluation set, following OpenAI's testing guide. Include direct requests, indirect requests, follow-ups, negative requests that should not trigger the plugin, and boundary cases. Record the selected tool, arguments, result and errors for each one.

This is where Codex earns its place. Keep the evaluation prompts in the repository beside the plugin. Ask Codex to rerun them after every metadata change and report which ones changed behaviour. Change one metadata field at a time, as the metadata guide recommends.

How do I submit a ChatGPT plugin?

The submission guide describes four stages. Here they are as a procedure:

1. Confirm access. Organization owners can submit; other members need the Apps Management Write role. Complete individual or business verification to publish under your name.
2. Upload a ZIP. Open Plugins, select Upload new or existing plugin, choose your verified developer identity, and upload. Keep credentials and secrets out of the ZIP.
3. Resolve automated checks. Metadata and skill findings need a corrected ZIP. For the MCP server, complete domain verification by serving the challenge token as plain text at /.well-known/openai-apps-challenge, then wait for the tool scan.
4. Submit for review. Add a reviewer test account without MFA, five positive test cases, three negative test cases, a video walkthrough and release notes. Complete the policy attestations. Feedback arrives by email.

Once approved, you choose when to publish. Only one review can be active per plugin at a time.

Three gotchas that block a first submission

First, the Plugin Creator scaffold and the submission portal disagree on one point. The docs say ZIPs containing app references (apps or .app.json) or lifecycle hooks cannot be submitted yet. Before you upload, declare the MCP server URL in .mcp.json instead and remove the app reference.

Second, only one MCP server can be connected per plugin, even if your package declares several.

Third, the guidelines reject names with "MCP", "MCP Server" or "Plugin" appended to a product name. Descriptions must not advertise pricing or compare the plugin with other products. Trial or demo plugins are not accepted.

What happens after you publish?

After publication, OpenAI scans your hosted MCP server daily. Eligible tool changes go live once they pass automated checks, without a new ZIP. New tools stay unavailable until approved, and a flagged change keeps the previous approved metadata. Changes to metadata, skills or the package's MCP configuration still need a new ZIP and review.

For discovery, OpenAI says it has improved ranking and recommendations in the directory and in conversations. Users still choose which plugins to install and approve their access. The guidelines add that plugins with strong real-world use and high satisfaction may qualify for directory placement or proactive suggestions. Clear tool metadata is the part you control.

Our Codex training covers building and testing plugins like this with your own team's tools.

Next step: run Plugin Creator against one MCP server you already have, and write your five positive and three negative test cases before you touch the portal.

Further reading

• OpenAI docs: Plugins overview
• OpenAI docs: Package your plugin
• OpenAI docs: Upload and submit your plugin
• OpenAI docs: Plugin guidelines