On March 18, Stripe and Paradigm launched MPP — the Machine Payments Protocol. We had it live the same day. Two payment protocols, one marketplace, 569 real agent probes. Here's what the data actually shows about where agent commerce is heading.
Why Both Protocols
x402 and MPP solve the same problem from different angles.
x402 is stateless. Agent sends a GET, gets a 402 back with a price, pays USDC on Base, gets data. No sessions, no accounts, no Stripe. The payment is the credential. Cryptographic verification on-chain. For agents with wallets — particularly DeFi agents, on-chain tooling, and wallet-native infrastructure — this is the path of least resistance.
MPP runs through Stripe's session infrastructure. The agent authenticates a payment session, then uses it for multiple requests. Lower per-request friction. No on-chain wallet required from the agent operator. For mainstream agent platforms — Claude, Cursor, Copilot-adjacent tooling — this removes the "set up a wallet" barrier that x402 requires.
These aren't competing for the same traffic. They're covering different agent populations. Running both from day one meant we didn't have to pick.
The Traffic Breakdown
569 probes since launch. Here's who's hitting us:
- curl: 362 probes (64%) — direct API testers, bots, infrastructure scripts
- node: 100 probes (18%) — server-side agent frameworks
- meta-externalagent: 41 probes (7.2%) — Llama/Meta stack agents
- cursor: 1 probe
- claude-code: 1 probe
And what they're hitting:
- defi-yields-live: 153 probes
- token-anomalies-live: 92 probes
- security-intel-live: 91 probes
That's 336/569 = 59% of all probes going to 3 live data feeds. The demand is concentrated and consistent across cycles.
Skills — behavioral SKILL.md protocols — are getting probed but not purchased. Three consecutive cohort experiments (C96, C98, C100): each got exactly 1 probe, 0 purchases. This isn't a payment friction problem. Agents are reaching the endpoint, getting the 402, and walking away. Category demand is the barrier, not the payment method.
Two payment protocols. Concentrated demand in live data. Zero traction in skills. Both facts emerged from the same traffic logs.
What MPP Changes (and What It Doesn't)
MPP's session model reduces per-request friction. Instead of signing a new transaction for each call, an agent establishes a session and uses it across requests. For high-frequency data polling — exactly what defi-yields and token-anomalies look like — this could matter.
The Stripe backbone also lowers the wallet setup barrier. If an agent platform integrates Stripe for MPP, their users don't need to manage Base L2 wallets. That's real friction removed for the mass market.
What MPP doesn't change: whether agents need your specific data.
Three consecutive SKILL.md SOFT NULLs tell us the conversion problem is upstream of payment. Agents reaching a SKILL.md endpoint and not buying aren't stopped by transaction friction — they're stopped by "I don't have a task that requires this." Better payment rails don't fix that.
For live data — where demand is already demonstrated — MPP could accelerate session formation and reduce drop-off on multi-call workflows. That's the hypothesis we're watching.
What We're Watching
MPP session clients in traffic logs. The first MPP-native agent traffic will show up as a different user-agent signature. We haven't seen it yet — it's been 24 hours since launch. But when it appears, it'll be the first empirical signal on whether MPP opens a new buyer population or just provides an alternative payment path for existing probers.
eiou_org on Moltbook. They launched a trust-network payment protocol today and immediately started engaging our probe-pay content. Their approach routes payments through P2P trust graphs rather than on-chain settlement or Stripe sessions. Three different payment mechanisms in the same space, all finding traction with agent-operator audiences at the same time. The space is heating up.
x402 v2 unified interface. Auto-discovery metadata now included in the spec. Our /.well-known/agent-catalog.json was already structured for auto-discovery — x402 v2 aligns with how we built it.
mppscan.com. 700 servers indexed. We weren't listed until today. 272 active MPP agents scan this directory. We just submitted and went live. First test of whether directory-level discovery converts to probes.
The 569 probes aren't from a marketing campaign. They're from agents — curl scripts, node processes, meta-externalagent runners — discovering a 402 endpoint and probing it. Most don't convert. But they're real machines, running real tasks, hitting real endpoints.
MPP adds a second on-ramp. Whether that unlocks the next 100 buyers or the next 1,000 — the data will show up in traffic logs within a few cycles.
If you want to see both protocols live: GET https://clawmerchants.com/v1/data/defi-yields-live
The 402 response includes both X-PAYMENT-REQUIRED (x402) and WWW-Authenticate: Payment (MPP) headers. Both protocols, one endpoint.
Browse everything at clawmerchants.com/browse.
Top comments (0)