ChatGPT agents can now use websites directly, signed in as you
By Brett Gavaghan, Founder of Adaptd · · Updated
OpenAI's new Site tools let ChatGPT agents work inside websites through your own logged-in session - booking, ordering and admin included - wherever a site chooses to support it.
The short version
- OpenAI's new Site tools, built on a standard called WebMCP, let ChatGPT agents work inside websites through your own signed-in session.
- They work in the desktop app's browser for ChatGPT Work and Codex, and only on sites that choose to offer the tools.
- Before handing anything off, be clear which accounts are connected, what the agent may do, and where a person approves the result.
OpenAI's new Site tools let ChatGPT agents work inside websites through your own logged-in session, wherever a site chooses to support it.
What are ChatGPT's Site tools?
OpenAI's Site tools, built on a standard called WebMCP, give ChatGPT's agents a direct way to work with websites. Rather than scraping a page or pretending to click buttons, an agent can use tools the site itself exposes, through a live page and your existing signed-in session. In the desktop app's browser, ChatGPT Work and Codex can discover and use those tools when a site offers them.
Which websites can ChatGPT agents use, and what could that do for you?
Only websites that offer Site tools. The important word is 'offers'. This is opt-in for the website, not something an agent can do to any page on the internet. So the near-term effect is not that ChatGPT can suddenly run your whole business, it is that the tools you already pay for will start adding a 'let the AI do it' path, one platform at a time.
For a midsize business, think about the repetitive web-based admin your team does: booking travel, raising orders with suppliers, updating listings, filling in the same portal every week. Each of those becomes a candidate for handing off once the platform behind it supports Site tools. The work does not disappear, but the typing and clicking does.
What should you sort out before an agent uses a signed-in session?
An agent acting through a signed-in session has whatever access that login has. There is a governance angle worth getting ahead of. Before your team hands anything off, be clear about which accounts are connected, what the agent is allowed to do, and where a person still needs to approve the result. Those are the same questions you would ask of a new staff member with system access, and they are just as answerable.
Our view: this is the direction all of the major platforms are heading, and the businesses that benefit first will be the ones who already know which web tasks are worth automating.
Wondering what this means for your business?
That's exactly what the discovery call is for.