Make your website usable by agents.

A W3C Community Group proposal that adds document.modelContext to the browser, so a web page can register tools an AI agent running in that browser can call directly — reusing your existing JavaScript, sharing the user's session, with the browser enforcing permissions. It's not a finished standard: Chrome and Edge ship it behind a flag or origin trial as of 2026.

It fetches your page's server-rendered HTML and looks for <form> elements, classifying each by its fields, action and submit-button text — search, login, contact, newsletter, checkout, or generic. The same limit as our agent-readiness checker applies: a form injected by client-side JS after hydration won't appear in a static fetch, because there's no headless browser behind this.

Review it first. Descriptions are inferred from field names and submit-button text — good enough to start from, not guaranteed accurate. Every registration is annotated with readOnlyHint and untrustedContentHint so an agent (and a human reviewing the code) can see at a glance whether a tool just reads or actually changes something, and the login/signup/checkout descriptions explicitly tell the model to confirm with the user first.

Because the API doesn't exist in most browsers yet. The generated snippet no-ops safely everywhere else instead of throwing.

The scan itself is — we fetch your page server-side to read its HTML, the same way the agent-readiness checker does. Nothing is stored.