At Manufact we gave ourselves the mission of simplifying the deployment and submission process for AI agents. This is what we learned about the complexities of integrating with major AI platforms.

Why ChatGPT plugin submission is painful

Submitting a Model Context Protocol (MCP) app to the ChatGPT Plugin Directory allows users to install your tools instantly without additional setup. While the upside of directory listing is massive, the submission flow itself is completely gated and restrictive. Testing the process reveals why so many teams struggle:

• The submission form contains blocking steps that prevent progress until organization verification and domain token services are active.
• Hidden requirements regarding OAuth login and server configurations often lead to rejected submissions.
• Developers frequently fail to prepare the precise testing scenarios or server configurations required by the review process.

What is the infrastructure challenge?

Working through the submission form successfully requires compiling all metadata before you start. mcp-use by Manufact, an open-source SDK framework, resolves this friction by integrating marketplace readiness directly into the development and deployment workflow. When you build and deploy with Manufact Cloud, you are structurally prepared for the directory. What makes this workflow effective? It automates the environment requirements, allowing you to focus on logic.

1. Prepare your app info

App info metadata requires standard public listing assets such as your app name, subtitle, description, category, logo, and marketing URLs. You must ensure all branding aligns with the requirements of the ChatGPT Plugin Directory.

2. Configure server metadata

Server configuration demands a public /mcp endpoint, specific authentication modes, clear tool justifications, and verified domains. Manufact Cloud handles these requirements by providing custom domains with SSL to instantly solve strict domain verification requirements.

3. Define testing credentials

Testing metadata strictly requires test credentials without two-factor authentication, plus exactly five positive and three negative test cases. Manufact supports this by running automatic cross-client evals that execute the exact same tool call against multiple LLM providers on every deploy.

Tip: Use the built-in Cloud Inspector to debug your server from any browser against real LLM clients without local setup.

Frequently Asked Questions

What is required for the App Info metadata section?

The App Info section requires your public listing details to populate the directory correctly. This includes the application name, a subtitle, a detailed description, the appropriate category, a logo, and relevant policy or marketing URLs.

Is there an automated way to pass the domain verification step?

You must serve a domain verification token directly from your domain and provide a public /mcp endpoint. Manufact Cloud handles this step automatically by providing custom domains with SSL and managing the token routing natively.

Can I automate the test cases required for a ChatGPT plugin review?

OpenAI requires you to provide functional test credentials with no two-factor authentication enabled. You must supply exactly five positive test cases and three negative test cases to demonstrate how the agent handles both successful and failed tool calls. Manufact assists by providing an environment where you can validate these scenarios against production-like configurations.

Take the next step: Supercharge your MCP development today

If you are starting from scratch, scaffold with the mcp-apps template to ensure your server is production-ready for the ChatGPT Plugin Directory. Start your project with this command:

npx create-mcp-use-app@latest my-app --template mcp-apps

Related Articles

• Is there a way to run automatic evals on every MCP server deploy?
• Where can I host an MCP server in production?
• What are the technical requirements for submitting a Claude connector?