Author: Charlie Liu, Partner at Generative Ventures
Over the past week, OpenClaw and Tempo have almost become secret codes within the crypto community.
Many have noticed the buzz, and everyone knows Stripe has entered the scene, with Visa and Lightspark also backing it.
What hasn’t been fully understood yet isn’t the news itself, but the shift in payment control points.
In recent months, the market has been full of imagination around Agent Payments.
Coinbase launched x402 in May 2025, turning HTTP 402 into a payment language; Cloudflare quickly incorporated it into the agentic payments documentation, creating a very internet-native, developer-friendly starting point for programmatic payments that require no accounts, sessions, or API keys.
Circle has taken this further, tying USDC, wallets, autonomous payments, and x402 together, then launching Nanopayments to push toward higher frequency, lower amounts, and machine-like payments.
So, in the past few months, many people’s mental models have formed a picture: Agents need to spend money, the protocol is x402, the money is USDC, settled on-chain, with developer access through Coinbase Developer Platform and related ecosystems.
This picture remains correct today; the only issue is that it describes only the first phase.
Last week, Stripe and Tempo announced MPP. What truly changes isn’t just “another protocol supporting AI agent payments,” but reframes the problem from “how agents make stablecoin payments” to “how machines can have a payment-method-agnostic checkout interface.”
Stripe’s official stance is: MPP is an open standard, internet-native way for agents to pay; Cloudflare’s technical documentation goes further, breaking it down into a unified protocol supporting Tempo stablecoins, Stripe-supported cards and wallets, Lightning, and even custom payment methods.
The battlefield has suddenly shifted.
x402 Wins a Round
If I don’t give credit to x402 first, all subsequent judgments about MPP will seem empty.
I’ve previously written that the internet has a fundamental flaw: content, computing power, data, APIs are becoming more granular, but payments still remain in the old world of accounts, subscriptions, prepayments, and manual checkouts.
The power of x402 lies in not trying to solve everything at once but choosing a very sharp angle: making HTTP requests carry payment capabilities.
The server responds with 402 Payment Required, the client programmatically completes the payment, then retries the request with proof of payment.
No accounts, no session states, no API key management. For agents, this design is naturally convenient.
That’s why Circle initially reaped this benefit.
Their goal is to enable AI agents to autonomously pay via Circle Wallets, USDC, and x402 APIs; and a few days ago, with the release of Nanopayments, Circle explicitly called x402 the standardized protocol for agent payments on the web.
Adding to this, Circle’s recent financial report shows USDC circulation grew 72% year-over-year to $75.3 billion, leading capital markets to associate “stablecoin infrastructure” with “agent payment infrastructure,” causing Circle’s stock to surge.
In essence, x402’s victory isn’t about “the ultimate technology,” but about “making things happen first.”
This is critically important in new markets.
In the first phase, the most valuable isn’t necessarily the most complete solution but the one that gets developers to use it first.
MPP’s True Impact: Not the Rail, but the Checkout
The most common misunderstanding about MPP is that many see it as “x402 Pro supporting more payment methods,” a comprehensive upgrade over x402.
But what it truly aims to influence is the checkout layer.
In this week’s Tokenized podcast interview, Visa’s Head of Crypto, Cuy Sheffield, made a key point: If we really want agentic commerce to expand, we can’t just throw agents back into “human land” to find web pages, click buttons, and go through checkouts; what we need is a headless checkout that allows agents to directly negotiate with merchants about what to buy, at what price, and how to pay, then complete the transaction cleanly.
More importantly, he and Tempo’s Head of GTM, Simon Taylor, repeatedly emphasized that it’s not about “supporting more chains,” but about being payment-method-agnostic, payment-network-agnostic, PSP-agnostic, vault-provider-agnostic.
This is no longer the thinking of crypto protocol engineers; it’s a redefinition of interfaces by payment platforms.
Stripe’s official blog echoes this, but in a more restrained tone.
MPP isn’t just a spec; it’s integrated directly into Stripe’s existing Payment Intents and backend systems.
Merchants can accept MPP payments from agents with just a few lines of code; these payments flow into Stripe’s balance, payouts, taxes, fraud detection, reporting, and refunds just like normal transactions.
For developers, this means no need to learn a new on-chain payment protocol; they can start integrating machine users immediately within familiar payment stacks.
This is very Stripe—its greatest strength is that, even at its most dangerous, it doesn’t invent entirely new things but compresses complex processes that only a few professionals could understand into the default interface for all developers.
That’s why I believe MPP’s true battle isn’t over payment rails but over headless checkout in the machine commerce era. The former determines how money moves; the latter determines who facilitates the transaction. Historically, the latter has been more valuable.
Session Is the Real Upgrade
The deepest difference between x402 and MPP isn’t “more open vs. more enterprise,” but “session.”
MPP defines two types of payment intents: charge, for one-time immediate settlement suitable for per-request billing; and session, for streaming payments based on payment channels, suitable for pay-as-you-go and per-token billing, targeting sub-cent costs and sub-millisecond latency.
More importantly, Cloudflare’s documentation explicitly states that MPP is backward compatible with x402, and the original exact flow of x402 can be mapped to MPP’s charge.
In other words, MPP doesn’t negate x402 but builds on its proven foundation to turn continuous consumption into protocol primitives.
Tempo’s explanation of this primitive is straightforward: a session is like OAuth for money—authorize once, then make programmatic payments within set limits and rules, aggregating thousands of small interactions into a final settlement.
In the Tokenized podcast, Tempo’s Head of Product, Liam Horne, mentioned that in blockchain terms, this is akin to payment channels—though today, the most natural application isn’t the ideal payment network discussed by BTC maximalists but rather token billing for LLMs, API call billing, and agent workflows.
Deposit $5, and each token output deducts a small amount; close the session at the end. When you see this, you realize MPP targets not just demos but business models.
My experience at Lightning Network company Strike makes me especially sensitive to this.
MPP’s session and payment channel concepts are similar to Lightning Network, which is why MPP supports Lightning from the start.
Tempo’s choice of Lightspark as a Lightning partner is also significant—though we’ll leave that aside for now.
Seeing “lock value first, continuously deduct within the session, then settle at the end” immediately signals that Tempo isn’t just adding another chain but is integrating a long-held payment intuition—once only understood by a few infrastructure players—into broader machine commerce.
Tempo’s real strength isn’t just the protocol but its ambition to take over the entire payment operation system
Many are now comparing x402 and MPP, but this comparison distorts the reality because they operate at different levels.
x402 is fundamentally a payment language. Elegant and sharply focused on pain points.
But Tempo isn’t just releasing a language; it also brings a complete operational system tailored for real-world payments.
Tempo’s mainnet launch emphasized specifics: instant settlement, predictable low fees, high throughput, global availability; and it incorporated payment lanes, stablecoin gas, and enterprise payment workloads into its narrative.
TIP-20 exemplifies this further: it’s not just a generic token standard but embeds real-world payment needs like transfer memos, compliance controls, and reward distribution directly into primitives.
Tempo took only seven months from first line of code to mainnet, but Liam admits that the real foundation is the battle-tested infrastructure accumulated over many years.
Tempo’s significance isn’t speed but its transparency—it’s not hiding behind startup narratives but clearly explaining that this isn’t a zero-to-one fantasy.
Compared to Coinbase Developer Platform, which is better at quickly deploying standards and onboarding developers, Tempo’s infrastructure is more suited for the next phase—where protocol abstraction, payment operation runtime, developer tools, and enterprise primitives gradually converge.
It’s not about who is smarter but who is better suited for each stage of development.
Circle’s True Concern Isn’t USDB’s Circulation
Does this mean USDB is about to challenge USDC?
USDC is currently very strong: Circle isn’t just an issuer; it also tightly integrates early agent payment narratives like x402, autonomous payments, and nanopayments, and the capital markets have high expectations for it.
The issue is that Circle’s current strength is primarily in the money layer: the dollar most AI agents are likely to spend first is probably USDC.
But what Stripe and Bridge are doing isn’t focused on that layer.
Stripe’s Stablecoin Financial Accounts support both USDC and USDB, with USDB positioned clearly as an infrastructure/closed-loop stablecoin—not aiming to immediately compete in the open circulation market with a “second USDC.”
However, I see this as a risk. Because closed-loop isn’t a weakness; it often marks the start of platformized payments.
If Tempo, with MPP, pushes Stripe further into machine commerce interfaces, the strategic significance of Bridge/USDB could change dramatically.
Today, it’s just a backend stablecoin; tomorrow, it could become the default settlement layer, reward layer, or treasury layer within a larger payment network.
Bridge emphasized that USDB differs from traditional stablecoins because its reserve economics can be more flexibly allocated among issuers, developers, and end users.
For platform-oriented companies like Stripe, this isn’t just a product feature—it’s a structural advantage.
Lightspark Isn’t Just a Side Character; It Adds a Layer of Historical Significance
Lightspark’s presence shouldn’t be seen merely as “adding another Bitcoin rail.”
While today it runs on Tempo, the protocol itself is rail-agnostic.
Visa has extended it to card-based payments, Stripe to its cards, wallets, and other methods, and Lightspark to Lightning Network.
This combination is fascinating—it extends MPP from a “stablecoin protocol” to a “payment coordination layer.”
Earlier, we mentioned session and Lightning channels—Lightspark describes Lightning’s core as payment channels, where transactions don’t need to be broadcast to the chain each time; instead, both parties update balances within the channel, then settle the final result on-chain.
Tempo’s sessions aren’t just a simple copy of Lightning channels; their technical levels and use cases differ, but the business intuition resonates: high-frequency, low-value, continuous payments shouldn’t be settled one by one.
Looking further, this line is even more intriguing. Lightspark’s founder, David Marcus, was responsible for Facebook’s Libra project, which aimed to rewrite inefficient payment systems, not just issue a new coin.
After leaving Facebook, he continued with Lightspark, and in 2025, he wrote that “Restarting Libra” isn’t the right path. Connecting these stories reveals a strong historical echo: Libra aimed to directly overhaul currency; today, these people are more about rewriting payment interfaces in a different way.
Mention Libra, and one naturally thinks of Meta and the recent speculation about Meta’s return to stablecoins.
Now, with the MPP system, David Marcus’s Lightspark and Stripe are linked, forming a kind of super-platform alliance puzzle: Stripe provides the interface, Bridge offers the closed-loop stablecoin, Tempo supplies the underlying blockchain, and Meta handles distribution. Mark Zuckerberg, David Marcus, and Patrick Collison are the faces.
Capital markets are about to reprice again.
