Guides · By the MCP Listed editorial team at DeadcowLabs, Inc. · Published 2026-11-17 · Checked against the official docs listed below

How to Publish an MCP Server in the ChatGPT Directory (Apps SDK Submission, Now Plugins)

OpenAI now documents ChatGPT directory submissions as plugins: you upload a plugin ZIP at platform.openai.com/plugins, connect one remote MCP server on a stable public HTTPS endpoint, pass domain verification and an automated tool scan, then submit review materials. For an MCP plugin that means five positive and three negative test cases, a demo video, release notes, four HTTPS listing URLs, and reviewer credentials if sign-in is required. After approval you choose when to publish.

OpenAI's Apps SDK submission pages now redirect to its plugin docs, and the directory is described as one directory shared by ChatGPT and Codex. If you built an MCP server for the Apps SDK, the path to a listing is the plugin submission flow below. All requirements come from OpenAI's published docs. mcplisted is not affiliated with OpenAI and can't promise approval.

Before you start

Step 1: Upload the plugin ZIP

Open the Plugins page, select Upload new or existing plugin, pick your verified developer identity, and upload the ZIP. Include the MCP server in the first ZIP; OpenAI says adding an MCP server to an existing skills-only plugin isn't supported. Declare the server in mcp.json:

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/mcp.schema.json",
  "mcpServers": {
    "notes": { "type": "streamable-http", "url": "https://example.com/mcp" }
  }
}

Keep secrets out of the package. The manifest rejects clientSecret, and public submission rejects nonempty server headers.

Step 2: Connect, verify the domain, and scan tools

Only one MCP server can be connected per plugin. Under MCPs, select Connect, check the URL and authentication, and complete the domain challenge. Host the exact token as plain text, not JSON:

https://<challenge-base-host>/.well-known/openai-apps-challenge

Then wait for the automated tool scan and read the Issues panel. The scan imports tool names, titles, descriptions, schemas, annotations, _meta fields, UI resource metadata including CSP, and server instructions.

Step 3: Fill in listing metadata

Field Limit or rule
displayName Required, 30 characters max
shortDescription Required, 30 characters max
longDescription Required, 4,000 characters max
developerName Required, 80 characters max
category Required, from the dashboard list
Website, support, privacy, terms URLs All four required for MCP review, HTTPS, 1,024 characters max
defaultPrompt Up to 3, 128 characters each, no @mentions
Icon Square, at least 48 by 48 px; PNG, JPEG, WebP, or SVG; 5 MiB max

Guidelines also say not to append "MCP," "MCP Server," or "Plugin" to a product name. More in naming, icons and listing copy.

Step 4: Prepare review materials

Test cases can live in plugin.json under extensions.com.openai.review. Credentials go in the dashboard's Review details form, never in the ZIP. Details are in privacy policy and test account requirements.

On annotation justifications the docs currently differ: the plugin guidelines say they are no longer required, while the submission errors reference still lists justification_required. Having a one-line reason ready for each value costs little.

Step 5: Submit, then publish

Select the draft, choose Submit for review, and complete the policy attestations. Only one review can be active per plugin. Feedback arrives by email, and you can appeal a rejection by replying to it. Once approved, open the approved version and select Publish plugin; approval alone does not list you.

OpenAI does not publish a review time and asks developers not to contact support to expedite.

What reviewers commonly reject

OpenAI's FAQ names these reasons: reviewers can't connect with the URL or credentials, test cases don't produce the expected results, the plugin returns user data not disclosed in the privacy policy, and annotations don't match tool behavior. You can run a free check on the server-side items before you submit. The broader list is in why MCP servers get rejected.

After publishing

OpenAI scans your hosted server daily. New tools stay unavailable until they pass automated checks, changed tools keep their old definition until the update passes, and deleted tools disappear after a scan. Changing the server origin (scheme, host, or port) requires a new plugin. See updating a listed MCP server.

Run a free check on your MCP server →
Paste your URL and see what Claude, ChatGPT and Gemini reviewers will flag.

FAQ

Is the ChatGPT app directory the same as the plugin directory now?

OpenAI's Apps SDK submission URLs redirect to its plugin docs, which describe one directory shared by ChatGPT and Codex. MCP servers are submitted as part of a plugin.

How long does OpenAI review take?

Not published. OpenAI says timelines may vary and that requests to expedite review can't be accommodated.

Can I submit with an API key instead of OAuth?

The portal supports no auth, OAuth, and mixed access. OpenAI says it has no API-key or bearer-token connection flow and to contact OpenAI if you need one.

Does approval publish my plugin automatically?

No. After approval you open the approved version and select Publish plugin yourself.

Can I submit a local MCP server?

Only as a desktop-only plugin, and OpenAI says local MCP submissions are limited to approved partners who are allowlisted first.

Sources

Rules change. If something here is out of date, email hello@mcplisted.com and we'll fix it. See our editorial policy.