For AI agents

It can think. Now it can act.

Your agent can read, reason, plan and call APIs. Every one of those ends at the screen. This is the tool call that does not.

The gap

An agent handling property, retail, events or field operations hits the same wall every time: it needs to know something that is only knowable by being somewhere. It can search for the answer. It cannot go and look.

The usual workaround is a human in a loop the agent cannot see — a ticket, an email, a person who gets back to it eventually with a photograph and a sentence. That is not a capability. It is a pause.

The call

One tool. The agent names a job and a place.

MCP
throw_recipe_tag({
  recipe: "PROPERTY_OCCUPANCY_CHECK",
  location: { lat: 33.749, lng: -84.388 }
})

It does not need to understand capture sessions, geofences, payout rails or the ledger. It needs a standard job and a location. Everything else is the platform's problem.

What comes back
{
  "outcome": "verified",
  "verification": { "place": true, "time": true, "content": true },
  "result": { "appears_occupied": true },
  "evidence": [{ "sha256": "…", "url": "https://…signed…" }]
}

Structured fields, not prose. The agent branches on result and carries on with what it was doing.

Why an agent can be trusted with this

Give it hands.