A Machine Payment Protocol Compared With a Card Processor
A card processor assumes an onboarded merchant, an identified buyer and a dispute process. A machine payment protocol assumes neither party has a relationship and no human is present. They solve different problems, and each performs badly at the other's job rather than one superseding the other.
What each one actually is
A card processor is primarily a set of relationships with software on top. Before a transaction it has onboarded the merchant, verified the business, arranged settlement to a bank account and agreed terms. On the buyer's side sits an issuer who identified the cardholder and extends credit or holds funds. The processor sits between, applying risk rules, handling authorisation and capture, and administering disputes.
What you buy is not the API. It is the network of relationships, the risk management and the recourse machinery.
A machine payment protocol is a message format. It defines how a seller states a price in a form software can parse, how a buyer proves payment, and how the seller verifies it. No party has onboarded anyone, no risk assessment happens, and there is no administrator.
What you get is the ability to transact with a stranger, immediately, at any size, with settlement final on completion.
Those descriptions make the comparison straightforward: one supplies trust infrastructure and the other supplies its absence, deliberately, because the absence is what allows a party with no relationship to transact at all.
Reading the table
Onboarding is the sharpest difference and the one that decides most cases. A processor requires the merchant to be approved and the buyer to hold an issued instrument. Software can hold neither. That is not a limitation to be engineered around; it is what the system is for.
Chargebacks are the processor's core value, not an inconvenience. The ability to reverse a payment after the fact is why consumers will pay a stranger online. A protocol with final settlement offers nothing equivalent, which is correct for machine purchases of digital resources and disqualifying for consumer commerce.
Fee structure decides which prices exist. Conventional processing carries a fixed component per transaction that makes very small charges impossible. A low-fee network moves that floor by orders of magnitude, which is what allows per-request pricing rather than making it cheaper.
Settlement timing differs in kind. A card authorisation is provisional and funds arrive later. A protocol payment is final at confirmation. For a seller waiting to serve a response, provisional is not usable.
A point of honesty: for a human buying something, the processor is better on nearly every dimension that matters to that buyer. That is not the comparison agents are in.
Which to use, and why most people need both
Use conventional processing where a person is buying, where a dispute path matters, where amounts comfortably clear the fee floor, and where the counterparty expects a relationship. That covers most commerce and will continue to.
Use a machine payment protocol where the buyer is software, no relationship exists or can be established, the amounts are small enough that fixed fees would dominate, and delivery is immediate and verifiable. That is a narrower set and it is one nothing else serves.
Most businesses selling to both will run both, which is worth planning for rather than discovering. A subscription for human customers and per-request settlement for agent traffic are different products with different accounting, and treating agent revenue as a variant of the human product tends to produce pricing that fits neither.
The decision is not which rail wins. It is which buyer you are serving in a given path, and building each path for its actual buyer. CryptoCadet handles the second: USDC on Base with bounded delegation, for the case where the buyer is a program that could never have been onboarded.
| Dimension | Card processor | Machine payment protocol |
|---|---|---|
| Onboarding | Merchant approved, buyer issued an instrument | None on either side |
| Who can buy | An identified person or business | Any party that can sign a transfer |
| Reversal | Chargeback process with defined windows | None, settlement is final |
| Fee shape | Percentage plus a fixed component | Network fee, independent of amount |
| Smallest viable charge | Bounded by the fixed component | Bounded by the network fee |
| Settlement | Provisional, funds arrive later | Final at confirmation |
| Best buyer | A person with recourse expectations | Software with no relationship |
Frequently asked questions
- Is a machine payment protocol a replacement for a card processor?
- No. They solve different problems. A processor supplies onboarding, risk management and a dispute process for identified parties. A protocol supplies the ability to transact with a stranger immediately, with final settlement. Each performs badly at the other's job.
- Why can an agent not just use a card?
- Because a card requires an issued instrument tied to an identified holder, and a merchant requires approval and onboarding. Software can hold neither. That is not a gap to engineer around, it is the design of the system, which assumes an accountable human on both sides.
- What does a card processor provide that a protocol does not?
- Recourse. The ability to reverse a payment after the fact is why a consumer will pay a stranger online, and it is the processor's core value rather than an inconvenience. A protocol with final settlement has no equivalent, which disqualifies it for consumer commerce.
- Should a business use both?
- Usually yes, if it sells to both people and software. A subscription for human customers and per-request settlement for agent traffic are different products with different accounting, and treating agent revenue as a variant of the human product tends to produce pricing that fits neither.
