WebMCP

Because of what it makes observable. Today an agent filling in your form on behalf of a real buyer and a script filling in your form to waste your time are the same event: a browser, driving a form, submitting values. Nothing in the request distinguishes them, which is why every defense in the anti-spam set is reduced to guessing from behaviour.

A callable surface separates them — not because agents are honest, but because the declared path is easier than the simulated one. An agent that can call a tool has no reason to drive a form, and something driving your form while a tool sits there unused has told you something without being asked.

• Nothing production-ready today. Watch the specification rather than building against it.
• The preparation that is not wasted either way: know which of your forms are the ones an assistant would reasonably be asked to fill in, and decide now whether you want that traffic.
• Record how each submission arrived, whatever the surface. If the distinction becomes measurable in 2027, the accounts that instrumented it in 2026 have a year of history and everyone else has none.

It will not, any more than crawlers reliably honour robots.txt. Identification happens when it is the path of least resistance, which means the design question is not “will agents behave” but “is the honest route cheaper than the dishonest one on my site.”

Build for that and good actors take the marked door for their own convenience. Build a marked door that is slower or more restricted than the form and nobody uses it, including the agents you wanted.