There’s already a schema.org spec that defines a JSON-LD structured data that you can embed on every of your product page to provide a machine readable interface of your product.
That`s is valid for search engines. But if JSON-LD was sufficient for agents, Google wouldn't have launched UCP (Universal Commerce Protocol) yesterday.
True. But extracting that metadata requires parsing the full DOM. CommerceTXT is for efficient discovery. Scan inventory cheaply first, then commit to the transaction.