In a recent LinkedIn post, Yonathan Cohen discusses how ChatGPT may recommend MCP plugins, arguing that conversion, repeat usage and retention shape which tools get surfaced. This article reports on Cohen’s post and analysis; it was not written by Cohen.
“the more users install your plugins and then use it frequently, the more likely we are to recommend it.”
Cohen attributes that explanation to Max Stoiber, whom he describes as working on ChatGPT plugins at OpenAI. The post presents three signals in a progression: users discover and install a plugin, use it repeatedly, and return the following week. Together, Cohen suggests, those behaviors could distinguish plugins that become regular tools from those tried only once.
Repeated work may support repeat usage
According to Cohen, builders should design an MCP around a task users perform every week. The recommendation follows from his emphasis on active usage and retention: a plugin connected to a recurring job may have more opportunities to become part of a user’s routine. Cohen does not provide data comparing different plugin use cases, but he frames frequency as central to the recommendation dynamic he describes.
“The catch: ChatGPT only recommends the top plugins. ‘Not the long tail.’”
That caveat gives Cohen’s advice a competitive dimension. If recommendations concentrate on a limited set of plugins, then broad availability alone may not ensure visibility. His post characterizes the challenge as earning enough installs and regular use to stand out among an expanding field.
Reduce friction at the start
Cohen also argues that first use should succeed “in one try.” In his framework, a smooth initial experience can help convert an install into meaningful usage. A user who cannot get the plugin working may not return, weakening the very repeat-use and retention signals Cohen identifies.
Bring in an initial user base
As Cohen notes, developers can encourage their existing users to install the plugin, creating an initial pool of users. He presents this as a way to build early usage, alongside making the plugin useful for recurring work and straightforward to try. The post does not describe a guaranteed path to recommendation; rather, it outlines behaviors Cohen believes may improve a plugin’s chances.
“SEO rewarded the best page. ChatGPT rewards the most used MCP.”
Cohen closes by contrasting search visibility with the usage-based model he sees emerging for MCP discovery. He says plugin submissions are up about 30 times this year, raising the stakes for developers competing for recommendation slots. His central point is that builders should focus on whether people install, repeatedly use and return to a plugin—not simply whether it exists.
📝 About This Content
This article is based on insights shared by Yonathan Cohen on LinkedIn.
📅 Originally posted on November 4, 2026 | View original post on LinkedIn →