How do you get an MCP app approved for ChatGPT and Claude?

Each platform runs its own directory and its own review, so treat them as two submissions. Read the current submission guidelines on each side before you start; the requirements change and an old blog post is not a safe guide.

Expect the review to look at the basics first: the server is reachable over HTTPS, sign-in works if you need accounts, and there is a privacy policy and a support contact. A reviewer who cannot get past login will not get to your features.

Tool design matters more than people expect. Clear tool names and descriptions, tools that only ask for the access they need, and actions that change data clearly separated from read-only ones all make a review easier and make the model use your server correctly.

Before submitting, test it the way a stranger would: a fresh account, the documented setup steps only, and a few ordinary requests. Fix whatever confuses the model there, because that is what the reviewer will see too.

Listings mentioned

• Creating Plugins For LLMs · skill by alpic-ai
  Covers the ChatGPT plugin lifecycle from build to deploy, with a reference for submitting to the directory.
• MCP builder · skill by anthropics
  A guide to designing clear, well-scoped MCP tools, which is most of what makes a server easy to review.

Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.