On March 18, Cuy Sheffield posted a single line to X: "Excited to share Visa CLI, the first experimental product from Visa Crypto Labs." The post hit 463K views in the past week. The product behind it: a command-line tool that lets AI agents execute secure Visa card payments directly from the terminal, no API keys, no pre-funded accounts, no human in the loop. An agent hits a paid endpoint, pays, gets the resource. One line of code. Access is closed beta, gated behind GitHub auth.

The supported use cases at launch are specific: image generation APIs, music generation endpoints, proprietary data feeds, market research behind paywalls, cloud compute. Anything an autonomous agent might need to purchase mid-workflow, Visa CLI handles it with a card payment instead of a credential. Sheffield's term for this category is "command line commerce," a future where agents transact autonomously rather than humans navigating checkout flows.

That same day, Tempo's mainnet went live alongside the Machine Payments Protocol. World shipped its x402 developer toolkit. Three competing agent payment standards active in the same week, all racing to become the default rail for autonomous AI spend. Visa's name is on all of them.

The MPP card spec is the real move

CLI is the consumer-facing hook. The infrastructure play shipped alongside it.

Visa released a card-based specification and SDK for MPP, the Machine Payments Protocol co-authored by Stripe and Tempo. Tempo is Paradigm's payments-focused blockchain, which raised at a $5B valuation from Thrive Capital in 2025. MPP defines a standard HTTP 402 flow: agent hits a service, service responds with a payment requirement, agent pays, service delivers. No accounts, no redirect, no checkout UI.

Visa's role is design partner. In practice: a card-native MPP extension so the protocol works across both stablecoin rails and Visa's global network, a full SDK for developers and merchants to build card-based transactions inside MPP flows, and access to Visa Intelligent Commerce and the Trusted Agent Protocol for tokenization and agent identity verification.

Lightspark extended MPP to Bitcoin Lightning the same week. The protocol now covers stablecoins via Tempo, cards via Stripe and Visa, and BTC via Lightning.

Visa's own language makes the strategy explicit: "While remaining protocol-agnostic, Visa built its card-based MPP specification to complement existing emerging agent payment standards, including protocols like x402." That sentence is the whole play.

The protocol war Visa is not picking sides in

Further reading of the 3 agentic payment protocols: The Race to Own Agent Payments: MPP vs. x402 vs. Google

Four serious contenders are competing for how AI agents pay online.

MPP launched with 100+ integrated services at mainnet. Rail-agnostic by design, supports stablecoins, cards, and Lightning. Stripe processed $1.9T in total payment volume in 2025, the distribution advantage MPP inherits directly.

x402 is Coinbase's stablecoin-only protocol, running USDC on Base. Artemis on-chain data puts it at around 131,000 daily transactions, average payment $0.20, roughly half testing activity. Coinbase claims 50M transactions through its Agentic Wallets infrastructure, though that figure includes the testing volume.

AP2 is Google's authorization and trust framework, chain-agnostic, compatible with both cards and stablecoins. The A2A x402 extension is production-ready for crypto; the card implementation is still maturing. ACP launched in ChatGPT's Instant Checkout in February, then OpenAI pivoted to an app-based model in March. Open standard with Stripe and Shopify support, but the direction shifted fast.

Visa has a formal named role on MPP only. But that framing misses what matters structurally. If an agent funds an x402 wallet with a Visa card, Visa collects upstream. If AP2 card implementations go live, Visa is already in the stack. The card network does not need to appear in a protocol spec to capture the settlement.

What Visa actually built

CLI and the MPP spec are the developer-facing products. What runs underneath is already live and larger than either.

In December 2025, Visa launched USDC settlement in the US. Cross River Bank and Lead Bank are settling Visa obligations in USDC over Solana, 7 days a week, at $3.5B annualized volume. Visa is also a design partner on Circle's Arc, a new L1 built to handle Visa-scale settlement volume, with Visa planning to run a validator node once Arc goes live.

Stack it together: the CLI brings card infrastructure into developer workflows where x402 and MPP are competing. The MPP card spec means Visa collects if MPP wins. The $3.5B USDC settlement layer means Visa is inside the stablecoin stack if any stablecoin rail scales. Arc gives Visa a validator stake in the settlement blockchain purpose-built for this traffic. Rubail Birwadker, Visa's global head of growth products, framed it plainly: "For these kinds of payments to scale, security isn't optional, it has to be built into every layer." That is how a 50-year-old card network makes itself indispensable to infrastructure that was supposed to route around it.

What to watch

The immediate signal is Visa CLI's developer adoption rate out of closed beta. If the GitHub waitlist converts and builders start shipping agents on card rails, Visa has a distribution wedge into the agentic stack that x402 and MPP cannot easily replicate. Neither Coinbase nor Stripe has 50 years of card acceptance infrastructure behind them.

The risk is fragmentation. If x402 or MPP gains enough volume on stablecoin rails that developers stop reaching for cards entirely, Visa's card-native CLI becomes a niche product rather than a foundation. The longer read is Arc. If Circle's L1 goes live with Visa running a validator, Visa is no longer just a card network with crypto exposure. That is a different company than the one CT thinks it is watching.

If you find this helpful, feel free to follow us for future updates. ❤

• X : @blockflow_news
• Telegram: t.me/blockflownews