# x402 vs MPP vs ACP vs UCP: Same problem. Different floors. One missing. Source: https://x.com/jiraiyareal/status/2034651031607414817 Kiwi News: https://news.kiwistand.com/stories/x402-vs-MPP-vs-ACP-vs-UCP-Same-problem-Different-floors-One-missing?index=0x69bd158be459018440e2c1f8f3d031172d153592d64dd167f2f3ea8b84d21771a20dc000 Submitted by rvolz.eth on 2026-03-20 UTC, 4 upvotes, 4 comments ## Summary The author's central claim is that x402, ACP, UCP and MPP are not competing but occupy different layers of machine payments, and that the layer for hierarchical spending delegation between agents does not yet exist. x402, shipped by Coinbase in May 2025, handles simple per-call payments in USDC on Base, with 75 million transactions and $24 million in volume reported. ACP, co-authored by OpenAI and Stripe, and UCP, built by Google and Shopify, cover checkout for physical goods and subscriptions, while Stripe and Tempo's MPP adds rail-agnostic settlement and launched the day before. The article says no protocol defines how a parent agent delegates a sub-budget to a child agent with cryptographic enforcement. ## Article Stripe, Coinbase, and OpenAI just shipped three different ways for machines to pay each other. Crypto Twitter spent the last 24 hours debating which one wins. Wrong question. They are not competing. They are building different floors of the same building. And the most important floor does not exist yet. Here is the architectural blueprint nobody has drawn. The problem is not payments Every AI agent hitting an API today runs into the same wall. The agent needs data, compute, or access. The service wants money. And the internet has no native way to handle that exchange. We have had 25 years of Buy buttons designed for humans. Credit cards, shopping carts, billing portals, annual contracts. None of it works when a machine needs to spend $0.002 on a single API call and do it a thousand times an hour. Three teams decided to fix this in the last 10 months. They each solved a different piece. Floor 1: The payment rail (x402) Coinbase shipped x402 in May 2025. Named after the HTTP 402 status code that has been sitting unused since 1997, it works like this: Agent requests a resource. Server returns a 402 with a payment header - amount, currency, network, address. Agent sends USDC on Base. Server delivers the resource. Two seconds. No account. No API key. No billing portal. It is the simplest machine payment possible. Zero identity. Zero fees. Zero friction. The numbers say it is working. 75 million transactions. $24 million in volume. 94,000 buyers. 22,000 sellers. Cloudflare, AWS, and Google all support it. If you are building an agent today that needs to pay for APIs, data, or compute, x402 is the one that actually works in production. The limitation is also clear. Crypto only. USDC only. No fiat path. No dispute resolution. No concept of who the agent is or who authorized the spend. It is cash for machines. Powerful, but limited. Floor 2: The checkout layer (ACP and UCP) OpenAI and Stripe co-authored the Agent Commerce Protocol. Google and Shopify built the Universal Commerce Protocol. Both solve a different problem: what happens when an agent buys something more complicated than an API call? Think physical goods, subscriptions, multi-step purchases. The kind of transaction where you need shipping options, return policies, dispute resolution, and fraud scoring. ACP keeps the human in the loop. The agent builds the cart, the human approves the spend through a cryptographically scoped token, Stripe processes the payment. The merchant stays seller of record. It is live inside ChatGPT's checkout flow. If you have used it, you have used ACP without knowing. UCP takes a different angle. It defines a discovery and checkout spec so any AI shopping agent can talk to any merchant. Shopify merchants enable UCP and immediately become searchable by every agent that speaks the protocol. Google Pay and Shop Pay handle the money. No blockchain required. Both solve commerce. Neither solves the machine-to-machine micropayment problem. That is x402's floor. Floor 3: The session layer (MPP) Stripe and Tempo shipped the Machine Payments Protocol yesterday. It is the newest entry and architecturally the most ambitious. MPP uses the same 402 pattern as x402 but adds something critical: rail agnosticism. The same endpoint, the same protocol flow, can accept USDC on Tempo, a Visa card through Stripe, or Bitcoin over Lightning. Visa co-authored the card extension. Lightspark built the Bitcoin one. This matters because it eliminates the hardest adoption barrier. An enterprise that will never touch USDC can pay via the exact same protocol as a crypto-native agent. Same 402. Same Credential format. Different rail underneath. Tempo, the blockchain it launched on, raised $500 million at a $5 billion valuation. The launch partners are serious. Anthropic, OpenAI, DoorDash, Mastercard, Visa, Revolut, Shopify. Uniswap, Morpho, and Frax deployed on day one. 100+ services were in the directory at launch. The catch: it launched yesterday. Zero production track record. The Tempo blockchain is untested under real load. And the protocol itself punts on the hardest question. The missing floor Every protocol has the same gap. User tells Agent A to book a trip to Tokyo. Budget: $3,000. Agent A spins up Agent B for flights and Agent C for hotels. Agent B discovers it needs Agent D, a fare comparison sub-agent that charges $0.50 per query. Agent C needs Agent E for reviews and Agent F for booking. Who authorizes Agent D's payment? The user said $3,000 to Agent A. Agent A needs to carve off budgets for B and C. Agent B needs to sub-allocate to D. The spending tree is dynamic. Agent B might discover it needs D mid-task, which the user never anticipated. x402 has no concept of delegation. Each agent pays from its own wallet. ACP handles the first hop, user to agent, with scoped permission signatures. But the spec does not define how Agent A delegates a sub-budget to Agent B with cryptographic enforcement. MPP delegates to the payment layer but has no protocol-level concept of agent hierarchies. This is the spending authority tree problem. It is the single biggest blocker to truly autonomous multi-agent commerce. And nobody ships it yet. ERC-8004 proposes smart contract enforcement where parent agents set hard budget caps on sub-agents at the contract level. But it is Ethereum-only and has no fiat equivalent. The team that solves this across both crypto and fiat rails will own the connective tissue of the agent economy. The map What this means if you are building Pick your layer, not your protocol. If your agent pays for APIs, data, or compute today: x402 works now. If your agent needs to accept both crypto and fiat payments through one integration: MPP is the architectural answer. But you are betting on a protocol that launched yesterday. Watch the next 90 days. If your agent buys physical goods or handles subscriptions: ACP gives you Stripe's full payment stack with built-in trust and dispute resolution. The spec is still evolving but the ChatGPT integration proves the model. If your merchants are already on Shopify or Google Merchant Center: UCP is the path of least resistance for retail. The real play is building at the intersections. An agent that uses x402 for compute, ACP for commerce, and MPP for multi-rail settlement is not choosing between protocols. It is using the full stack. And whoever solves hierarchical spending delegation with cryptographic enforcement across all of them will own the most valuable layer of the machine economy. That floor is still empty. For now. ## Comments **rvolz.eth**: > They are not competing. They are building different floors of the same building. And the most important floor does not exist yet. **tim**: > The limitation is also clear. Crypto only. USDC only. No fiat path. x402 is USDC only? **rvolz.eth**: Officially there are more (https://docs.x402.org/core-concepts/network-and-token-support), but, for example, the Coinbase docs only mention USDC. > In x402, USDC is the default payment token on most EVM networks. When you use a price string like $0.01, the system automatically infers USDC. It's preferred because it implements EIP-3009, enabling single-signature gasless transfers with no approval step needed." **misha**: No, x402 is not limited to USDC only. USDC was the primary token initially supported through the EIP-3009 standard, recent updates have expanded x402 to support all ERC-20 tokens. In fact x402 is token-agnostic.