Yep, you basically have that right. Using AI in your projects in order to help you build, or train your own agents is OK, as long as you’re using it on your own stuff.
You can’t train machine learning or AI models on Unity itself or on data derived from it without our authorization, you can’t point automated scrapers or bots at Unity’s APIs or the Asset Store, and AI agents, LLMs, and MCP clients/servers need to connect through Unity’s Authorized Agentic Access framework rather than poking at the platform directly. 3rd party MCPs are and will continue to be allowed through authorized channels (including, the asset store, verified partners program, or where they are otherwise indicated that they meet Unity Core Standards), or if not on a banned list on documentation.
So, while we’re not trying to prevent things like 3rd party MCPs (that would be a game of whack-a-mole), we have the language as it’s written here in order to help us maintain the integrity of the platform and assist us in ensuring that third parties interact with our software (and your project) in acceptable ways. This is not only to protect our IP but also the IP and data of our users (aka: we need to be able to protect everyone from possible “bad actors”)
(full transparency, I’ve copy pasted this from other posts I’ve made)