agentleFS
saurabhyadavzsaurabhyadavz/publish-mcp-server-claude-chatgpt/skills/submit-to-chatgpt/SKILL.md

submit-to-chatgpt

Skill1 starsChanged 9 days agoMITView on GitHub

Walk through publishing an MCP server as a ChatGPT plugin at platform.openai.com/plugins, step by step - testing it first as a custom MCP server in ChatGPT, uploading the ZIP, connecting the server, the domain-verification token at /.well-known/openai-apps-challenge, reviewer credentials, test cases, the demo video, reading the findings, and submitting. Use when someone is on the OpenAI plugins portal, asks where to find the domain token or the Submit button, gets a portal finding or "needs further review" message, needs a demo video for review, or wants to know if they can submit yet.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After every deploy: open the plugin and click Refresh (ChatGPT caches tools and card HTML).

    line 22
  2. Very specific:

    Run every review test case.

    line 23
  3. Very specific:

    Check it with curl before clicking Connect/Verify.

    line 43

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Submitting to the ChatGPT plugin directory

Two different sites: chatgpt.com/plugins is where you use and test plugins; platform.openai.com/plugins is where you submit one. "Upload plugin archive" on chatgpt.com only installs it for yourself. Full portal walkthrough: references/portal.md.

0. Test in ChatGPT first

chatgpt.com/plugins → Add → Add custom MCP server → name, the MCP URL (ending in /mcp, nothing after it), OAuth → "I understand and want to continue" → Create as a plugin → sign in on your page as the reviewer → Install. In a new chat type @ and pick the plugin.

  • After every deploy: open the plugin and click Refresh (ChatGPT caches tools and card HTML).
  • Run every review test case. If ChatGPT answers without calling your tool, fix the server instructions (see mcp-server-for-plugins), not the prompt.
  • Watch cards: "Opened <card>" with an empty box, or "timed out", means the card's start-up or data pickup is wrong (see mcp-app-cards).

1. Upload the ZIP

platform.openai.com/plugins → Upload new or existing plugin → the ZIP from plugin-package (run its checker first). Fixing the package later: Upload new version on the plugin's page. A deleted test plugin in ChatGPT doesn't touch the draft here.

2. Connect the server and verify the domain

MCPs tab → Connect (or Reconnect). The panel shows the URL, OAuth, the redirect URI (https://chatgpt.com/connector/oauth/<id>) and, further down, the domain-verification token. The status badge itself isn't clickable.

  • Serve the token as plain text, nothing else, at https://<your MCP host>/.well-known/openai-apps-challenge (a route that reads it from an environment variable is simplest). A new draft can get a new token.
  • Check it with curl before clicking Connect/Verify. Paste the real token, not a placeholder: the endpoint will happily serve the word "TOKEN".
  • Sign in as the reviewer when asked. "Authorized" and a green domain badge mean this step is done.

3. Review information (five steps in a side panel)

  1. Countries and translations: from the ZIP.
  2. Reviewer credentials: username and password; Login URL = the sign-in page your OAuth flow opens; no tenant unless you have workspaces; notes with the exact steps a stranger follows. You type the password; never paste it anywhere an agent or a log can read it.
  3. Test cases and 4. Negative test cases: from the ZIP (read-only here).
  4. Supporting content: the demo video URL (unlisted YouTube or a Drive link anyone can view) and the release notes.

4. Read the findings, then submit

  • "This new tool definition needs further review before it can be published": not an error. A person will look at that tool (often tools that serve news or third-party content). Be ready to say where the content comes from. Check the Blocking tab; if it's empty you can submit.
  • Contrast, auth-settings or schema findings: fix the package and upload a new version.
  • Click Rescan after a deploy so the checks run against the current server.
  • Submit for review becomes usable when nothing says "incomplete". Publish only after approval.

The demo video

Show exactly what reviewers will do: connect and sign in, all five test prompts in a fresh chat with each card fully drawn, one refused prompt. Before recording, copy the password first and close the password manager, notes and terminal. If they appear, cut them; the video tooling in your agent or any editor can trim. A test case that visibly fails in the video invites a rejection: re-record that clip instead of shipping it.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.