Authorize the action the protocol will carry.
Sandbox keys in five minutes, no card, running the same policy engine as live mode. Live tiers from $49/mo.
The control layer ยท Versus agent payment protocols
The landscape
The gap
It proves this agent was authorized for this checkout. It does not know that the same agent already spent its month by lunchtime, because nothing in the protocol carries a running total.
A token scoped to one merchant and one window is a good control for that merchant. Your agent has an OpenAI bill, a data vendor, and a freight broker, and no rail sees the other two.
The model call, the enrichment API, the search credit. None of it is a card transaction, so none of it is in scope for a checkout protocol, and it is often the larger number.
A protocol can prove the transaction was authorized. Answering which policy permitted it, with what headroom left, is a different record, and it has to be written by whatever made the decision.
Adopting one
Adopting AP2 or ACP is not a decision against Kordio, and if a rail you use ships one you should take it. The protocol settles identity and terms at the moment of checkout. Kordio settles whether the spend was ever supposed to happen, and it does that before the protocol is invoked.
Kordio does not implement these protocols today. It runs in front of whatever rail you already use and returns a decision your own executor acts on, so the order is Kordio first, then the rail.
Every decision is recorded in an append-only audit trail, linked by trace id, on every tier. See what comes back
{
"decision": "requires_approval",
"rule": "approval_threshold.freight",
"remaining_cents": 152000,
"cosignature": "eyJhbGciOiJFUzI1NiJ9..."
}Overlap questions
Yes, and they do different jobs. A protocol proves the agent is who it says at checkout. Kordio decides whether that spend is inside the budget you set, then signs the answer.
No. Kordio runs in front of whatever rail you already use, and returns a decision your own executor acts on. If you adopt a protocol, Kordio still answers first.
No. MCP standardizes how an agent reaches a tool. It is on this page because paid tool calls are agent spend too, and most spend controls never see them.
Their limits scope one token on one rail. Kordio holds one budget across runs, spanning payments and tool calls together, so spend on one consumes headroom for the other.