When I first started tracking how AI agents would pay for things, the card networks felt like spectators. By mid-2026 that has flipped completely: Mastercard Agent Pay and Visa's Trusted Agent work are two of the most consequential pieces of agentic-commerce infrastructure in the market. Both wrap the same underlying idea — a machine that can prove who it acts for and what it was told to buy — but they package it differently. In this guide I walk through how each approach works, put them side by side, and show where crypto rails like x402 and stablecoins actually plug in. I've built and tested small agent-checkout flows against these APIs, and I'll flag where the marketing gets ahead of what you can ship today.
What Agentic Tokens Actually Are
An agentic token is a payment credential minted specifically for a software agent, not for a human tapping a phone. On the card side it's an extension of the network tokenization systems that already replace your real card number at checkout — Mastercard builds its Agentic Tokens on top of the Mastercard Digital Enablement Service (MDES), the same tokenization plumbing that powers device wallets today.
The important shift is what rides with the token. A normal network token just says "this is a stand-in for card X." An agentic token adds verifiable context about authorization: which agent is presenting it, on whose behalf, and within what limits. This is where Google's Agent Payments Protocol (AP2) matters. AP2 defines three nested mandates — Intent, Cart, and Payment — each signed as a verifiable credential. The Intent mandate captures what you asked for ("book a flight under $400"), the Cart mandate pins the exact items and price, and the Payment mandate authorizes settlement. Both card networks have converged on wrapping their agentic tokens inside these AP2 mandate envelopes so an agent's action carries a cryptographic paper trail.
Mastercard Agent Pay: How It Works
Mastercard Agent Pay was first announced on April 29, 2025, and has matured through 2026 into a framework where verified AI agents transact using Agentic Tokens. Mastercard joined AP2 as a launch partner in September 2025, and the practical result is that an AP2-compliant agent can pay over Mastercard rails with the AP2 mandate envelope wrapping a Mastercard Agentic Token.
By mid-2026 the direction is to expose AP2 Intent and Cart mandates natively inside the Agentic Token provisioning API. In plain terms: when your agent requests a token, it hands over the signed Intent and Cart credentials as part of the same call, rather than bolting authorization on as a separate step. That tightens the link between "what the user consented to" and "what got charged," which is the whole point of the mandate model — it gives the merchant and issuer an evidence layer if a purchase is later disputed. The flow leans on existing card acceptance, so the merchant side largely looks like a tokenized card transaction with extra credential data attached.
Visa Trusted Agent Protocol and Intelligent Commerce
Visa's strategy is broader and more integration-focused. On April 8, 2026, Visa announced Intelligent Commerce Connect, described as a single-integration on-ramp to agentic commerce. Rather than backing one protocol, Connect explicitly supports Visa's own Trusted Agent Protocol alongside the Machine Payments Protocol, the Agentic Commerce Protocol, and the Universal Commerce Protocol.
The Trusted Agent Protocol focuses on the identity and trust problem: letting a merchant confirm that an incoming agent is legitimate and is acting inside a delegated authority, so the merchant can treat agent traffic differently from bots or fraud. The mid-2026 move is to unify Trusted Agent Protocol and Intelligent Commerce under a single SDK with AP2 compatibility by default. For developers that means one integration surface that speaks AP2 mandates out of the box while still leaving room to route through the other protocols Connect supports. Visa has also been aggressive on settlement: its stablecoin settlement pilot reached a roughly $7 billion annualized run rate by April 2026, running across nine blockchains and 130-plus stablecoin-linked card programs in 50-plus countries — which matters for the crypto section below.
Mastercard vs Visa: Side by Side
Here's how I'd summarize the two approaches as they stand in mid-2026. Treat this as a snapshot — both are moving fast and telegraphing feature parity.
| Dimension |
Mastercard Agent Pay |
Visa Trusted Agent / Intelligent Commerce |
| Core credential |
Agentic Tokens (built on MDES) |
Tokenized Visa credentials via Intelligent Commerce |
| AP2 posture |
Launch partner (Sept 2025); Intent/Cart mandates native in Agentic Token API |
AP2 compatibility by default in the unified SDK |
| Primary emphasis |
Token provisioning + verifiable intent |
Agent identity/trust + single integration on-ramp |
| Integration surface |
Agentic Token provisioning API |
Intelligent Commerce Connect (one SDK, multi-protocol) |
| Other protocols supported |
AP2-centric |
Trusted Agent, Machine Payments, Agentic Commerce, Universal Commerce |
| Stablecoin settlement |
Participating in x402 standardization |
Live pilot ~$7B annualized, 9 chains, 130+ programs |
| Best fit |
Teams wanting tight mandate-to-token binding |
Teams wanting one SDK across many agent protocols |
The honest takeaway: neither is strictly "better." Mastercard optimizes for a clean, verifiable line from user consent to charge. Visa optimizes for breadth — one integration that tolerates whichever agent protocol shows up. If you only need card rails and strong intent evidence, Mastercard's model is tidy. If you're integrating many agent sources, Visa's aggregation is attractive.
Where Crypto Rails (x402 and Stablecoins) Fit
This is the part people miss: the card networks and crypto rails are converging, not competing. On July 14, 2026, the Linux Foundation launched the x402 Foundation with 40 members — including Visa, Mastercard, Stripe, AWS, Anthropic, and Circle — to steward the HTTP 402-based protocol for autonomous stablecoin payments.
Architecturally, x402 and AP2 are complementary. AP2 handles the authorization and mandate layer (who consented to what); x402 handles the on-chain settlement leg. Google's own documentation frames x402 as the stablecoin settlement extension for AP2, and the A2A x402 extension was built jointly with Coinbase and the Ethereum Foundation. Settlement typically uses USDC with EIP-3009, which enables gasless authorization signatures so the agent doesn't need to pre-fund gas. So the same AP2 Intent/Cart/Payment mandate that fronts a Mastercard Agentic Token can, instead, front an on-chain USDC payment.
Reality check on scale: x402 processed roughly 75 million transactions but only about $24 million in value over a recent 30-day window — mostly sub-dollar, machine-to-machine payments like per-API-call billing. Card rails still dominate real consumer spend. The likely 2026 pattern is both: mandate-authorized card tokens for retail-style purchases, and x402 stablecoin settlement for high-frequency micropayments where card economics don't work.
What This Means If You're Building Agents
If you're wiring up an agent to spend money today, my practical read is this. Design against AP2 mandates first, because that layer is now the common denominator across Mastercard, Visa, and the x402 crypto path — it keeps your options open. Pick the card network by integration fit, not hype: Visa's unified SDK is convenient if you juggle multiple agent protocols; Mastercard's native mandate-in-token flow is cleaner if you want the tightest consent-to-charge binding. Add crypto settlement via x402 only where micropayments or programmatic API billing genuinely need it. And keep spend limits, allowlists, and human-in-the-loop confirmation on top of any of these rails — the payment protocol proves intent, but it doesn't stop a compromised or misled agent from acting inside its granted authority.
Frequently Asked Questions
Is Mastercard Agent Pay the same as Visa Trusted Agent?
No. They solve overlapping problems with different emphasis. Mastercard Agent Pay centers on Agentic Tokens with AP2 Intent/Cart mandates baked into token provisioning. Visa's Trusted Agent Protocol centers on agent identity and trust, and by mid-2026 is folded into a single Intelligent Commerce SDK that speaks AP2 by default.
Do I have to choose between card rails and crypto?
Not really. AP2 is designed so the same mandate structure can settle over a card token or over x402 stablecoins. Many teams will use cards for consumer purchases and x402/USDC for high-frequency machine-to-machine micropayments.
Are these live and production-ready in 2026?
Partially. The frameworks, SDKs, and standards (AP2, the x402 Foundation, Visa's Connect) are real and shipping, but availability, supported regions, and merchant coverage vary. Treat everything here as fast-moving and verify current rollout status before you commit an architecture.
What are agentic tokens in one sentence?
They're payment credentials minted for an AI agent that carry verifiable proof of who authorized the purchase and within what limits — network tokenization plus a signed consent trail.
Conclusion
Mastercard Agent Pay and Visa's Trusted Agent work are two lenses on the same 2026 reality: agents need credentials that prove delegated intent, and AP2 mandates have become the shared language for that. Mastercard binds those mandates tightly into token provisioning; Visa aggregates protocols behind one SDK. Crypto rails aren't the rival here — x402 and stablecoins slot underneath AP2 as the on-chain settlement option, standardized by a foundation that both card networks helped launch. My advice: build to the mandate layer, choose the network by integration fit, and layer your own spend controls on top. That keeps you portable as this space keeps shifting month to month.