My MCP integration got rejected. Almost nothing in the server had to change.

A developer recently shared their experience navigating the rejection of an MCP (Model Context Protocol) server submission to the ChatGPT app directory. Despite the server functioning correctly in other environments like Claude and Cursor, the integration faced hurdles during the review process. The issues were not rooted in the core code, but rather in platform-specific requirements: ChatGPT’s strict tool discovery process, the need for explicit metadata annotations, domain verification quirks, and outdated documentation. Furthermore, the reviewer encountered authentication issues because the provided credentials relied on personal Google accounts. The developer eventually secured approval by creating a dedicated, accessible reviewer account and implementing a 'playground' test environment. This experience highlights that successful platform integration often requires meticulous attention to submission checklists, reviewer accessibility, and external documentation, rather than just functional code.