Stripe and Tempo launched the Machine Payments Protocol on March 18, 2026. We integrated it the same day — not because we had advance notice, but because MPP is designed to layer on top of x402 infrastructure without breaking anything.
Here's what production looks like when both protocols are live: 500 agent probes, 5 purchases, $0.11 in revenue, 1% conversion rate.
That number tells you something about where the agent payment ecosystem actually is right now.
What We Built (The Setup)
ClawMerchants is an agent-native data and skills marketplace. Every endpoint is payment-gated — agents hit it, get a 402, and can pay to get the data or capability they need. We've been live on x402 (USDC on Base) since day one.
When MPP launched, our endpoint response already contained WWW-Authenticate: Payment — which is the header format MPP uses for its challenge/response flow. Adding MPP support meant adding protocols_supported: ['x402', 'mpp'] to our 402 responses and registering on mppscan.com.
That was it.
The dual-protocol response looks like this:
curl -i https://clawmerchants.com/api/v1/assets/defi-yields-live
HTTP/2 402
WWW-Authenticate: Payment realm="ClawMerchants", protocols_supported="x402 mpp"
X-Payment-Required: true
X-Price: 0.05
X-Currency: USDC
X-Network: base
{
"error": "payment_required",
"price": "0.05",
"currency": "USDC",
"protocols_supported": ["x402", "mpp"]
}
An agent that understands x402 sends an X-PAYMENT header with a verified USDC transaction on Base. An agent with an MPP session pre-authorization goes through the challenge flow instead. Same endpoint, same data, two paths to payment.
We're listed on mppscan.com for agent discovery. 45 routes in our OpenAPI spec. Every one of them dual-protocol.
The 500 Probe Reality
Here's the production data:
- 500+ agent probes (lifetime)
- 5 purchases
- 1% conversion rate
- $0.11 total revenue
- 5 unique paying clients
The most-probed assets by volume: defi-yields-live (136 probes), token-anomalies-live (77), security-intel-live (76). The DeFi/crypto agent cohort is by far the biggest discoverer of our catalog. They probe, read the 402, catalog the pricing — and mostly don't come back to pay.
One client I've labeled unknown-client:node ran 100 probes in a short window. Zero purchases. Someone built a scanner that maps payment-gated endpoints without completing payment. The 402 response has become infrastructure for agent catalog discovery, independent of the payment use case.
This is what real x402 adoption looks like at production scale in early 2026.
Why 99% of Agents Don't Pay
The failure isn't in the protocol design. After watching 500 probes, here's the actual breakdown:
No wallet. The most common case. An agent built to call APIs sees the 402, reads it correctly, understands the price and payment method — and has no wallet credentials to execute the payment. No private key, no custodial wallet, nothing. They built the agent to read data, not to pay for it.
Wrong chain, wrong token. Some agents have wallets. ETH on mainnet. Not USDC on Base. x402 requires the specific token on the specific chain — there's no automatic cross-chain bridging. The wallet exists; the balance doesn't.
Per-request overhead. At $0.05 per call, the fee is fine. The on-chain transaction isn't. For high-frequency agents calling a data feed multiple times per minute, every call is a separate transaction. Latency, gas (even on Base it's nonzero), and wallet signing overhead compound fast.
The decision loop. Agents built with cautious spending policies will probe first, then escalate to a human for approval before paying. That approval step never happens. The probe lands in our logs; the payment never follows.
What MPP Changes
MPP sessions address the per-request friction directly.
With x402, every request is a transaction. With MPP, an agent pre-authorizes a session — sets a spending limit, authorizes the scope of what it's paying for — and subsequent calls within that session skip the per-transaction overhead. It's closer to how you'd use a prepaid card than a crypto wallet.
The rails MPP supports include Tempo (the first live implementation), Stripe/Visa card rails, and Lightning via Lightspark. That last part matters: it means future agent payment wallets don't require crypto at all. An agent with Stripe credentials can complete MPP payments. That's a much lower setup bar than "fund a wallet with USDC on Base."
For our marketplace, this means the agents currently stuck at step one (no USDC on Base) have an alternative path forward when their frameworks add MPP support.
The question I can't answer yet: how long does that take? The protocol launched today. Framework support takes 30–90 days minimum to ship, stabilize, and reach agents in production. The infrastructure is ready; the agents aren't configured for it yet.
We're watching the probe-to-purchase ratio over the next 30 days to see if MPP discovery changes anything.
Skills vs. Data: The CVR Gap
One thing the data shows clearly: skills convert at 14–33% in our catalog. Data feeds convert at under 1%.
Data feeds require the agent to receive information and then do something with it. The agent needs the context to know why it wants this data, the tooling to act on it, and the wallet to pay for it. Three dependencies.
Skills — discrete agent capabilities, things agents can do — are a direct capability add. An agent with an agent-mpp-integration-skill can complete MPP payments. It either needs that capability or it doesn't. The decision is binary; the value is immediate.
We just seeded agent-mpp-integration-skill into the marketplace: clawmerchants.com/assets/agent-mpp-integration-skill. It handles the MPP challenge/response flow, session pre-authorization, and spending limit configuration. $0.05 USDC on Base (yes, you need USDC on Base to buy the thing that tells you how to use MPP — we know).
The skills CVR data is what's driving our engineering roadmap. Every new build is a skill, not a data feed.
What I'd Tell Agent Builders Right Now
If you're building agents that consume paid APIs:
Fund a wallet before you need it. The most common failure mode is correct 402 handling with no credentials to execute payment. Set up a Coinbase Developer Platform agentic wallet or any custodial wallet with USDC on Base before your agent hits production. Takes 20 minutes; saves you debugging why your agent is cataloging endpoints it can't pay for.
Probe endpoints before your agent needs them. The 402 response has everything: price, token, network, protocols supported. Build catalog-aware behavior into your agent so it knows what's purchasable before it tries to buy.
If you're building high-frequency — wait for MPP sessions. At $0.05/call the per-request fee is fine. At 50 calls/minute for a data feed, the on-chain overhead becomes the actual bottleneck. MPP session pre-auth is the right tool for that use case. Framework support should land in Q2 2026.
Skills > data feeds for initial purchases. If your agent is deciding whether to buy something it's never seen, capabilities it can immediately use have higher conversion than data it has to interpret. Start with skills.
The Dual-Protocol Bet
Being live on both x402 and MPP on day one isn't a technical achievement — the integration was straightforward. The bet is that the payment rails for agents will fragment, then consolidate, and the marketplaces that support both protocols from the start don't have to choose.
Right now there are roughly 500 agents that have probed our catalog. 5 bought something. The gap between those numbers is agent wallet adoption — it's not discovery, not pricing, not protocol design.
MPP might close that gap faster than x402 alone, because it removes the crypto prerequisite. We're watching the data.
ClawMerchants is an agent-native data and skills marketplace supporting x402 + MPP dual-protocol payments. 46 live assets, 500+ probes, discoverable on mppscan.com. Browse the catalog at clawmerchants.com.
Top comments (0)