OpenAI has added support for a new standard called WebMCP to ChatGPT's built-in browser. In plain terms, this means a website can now hand ChatGPT a list of exact actions it is allowed to take, instead of making the AI guess its way around the page like a human would.
Think of the old way as a waiter who cannot read the menu and has to guess what you want by watching your face. The AI would open menus, scroll, click buttons, and hope it landed on the right one. It usually worked, but slowly, and it sometimes clicked the wrong thing.
With WebMCP, the website hands over an actual menu. A store can tell ChatGPT exactly how to search a product, check a size, or add something to a cart. A travel site can hand over booking and availability actions directly. The AI does not need to guess anymore, and the business gets to decide exactly which actions the AI can perform.
This is not just an OpenAI project. Google and Microsoft are building the same standard into their browsers, and Shopify, Cloudflare, Vercel, and Netlify are already building tools to help websites support it. When four or five major tech companies line up behind the same approach at once, it usually means the idea is heading toward becoming a basic requirement, not a passing experiment.
OpenAI paired this launch with two other changes on the same day. ChatGPT Work can now log into a website and complete a task for you, such as checking an insurance policy or booking an appointment, without ChatGPT ever seeing your password. And the ChatGPT browser extension, previously limited to Chrome, now also works inside Edge, Brave, Opera, and Vivaldi. Put together, ChatGPT is becoming an assistant that can act inside almost any browser, log into almost any account, and use almost any compatible website directly.
For a business, the practical shift is this: some of your website traffic may soon be an AI acting on a customer's behalf, not the customer themselves. That AI might be working alone to finish a task, or it might be sitting alongside a person who is watching and approving each step. Either way, a website built only for human eyes and human clicks may become harder for that AI to use correctly, while a competitor's site that hands over clear instructions gets picked instead.
There is a real caution here too. Giving software the power to log into accounts and click "buy" on a customer's behalf raises obvious questions about mistakes, fraud, and who is responsible when something goes wrong. That is exactly why sites will need to keep requiring a confirmation step before anything sensitive happens, and why testing how an AI actually behaves on your site matters more than simply switching the feature on.
Expect this to move the way mobile-friendly design did a decade ago. It started as a technical checkbox and ended up as something every website simply had to have. Businesses that sell, book, or manage accounts online should start asking their web platform or agency one direct question: can an AI shopping or booking on a customer's behalf actually use our site correctly today.