BTC —◆ ETH —◆ SOL —◆ STABLECOIN FLOAT $313.58B◆ NATIVE USDC ON BASE $4.31B◆ DEFI TVL $95.61B◆ THE PUBLIC LEDGER OF MACHINE MONEY — AGENTICFINANCEGRAPH.COM◆

Home/Research/x402 and HTTP 402

Payments · the rail

A status code sat empty for thirty years. Then machines needed it.

Cards have a thirty-cent floor and agents buy three-cent API calls. That single arithmetic fact is why a reserved HTTP status code became a payment rail with Visa and Mastercard in the room.

Agentic Finance Graph · Published 2026-09-12 · Dates from public launch posts and foundation announcements · ~1,500 words

Direct answer x402 is a payment protocol that gives HTTP status code 402 a meaning after thirty years of it being reserved and unused. Coinbase published it on 6 May 2025. A server answers a request with 402 and its payment requirements, the client signs a stablecoin authorisation, a facilitator verifies and settles, and the resource is returned — with no account and no session. The Linux Foundation has stewarded it since 2026, with forty member organisations.

Everything interesting about x402 follows from one decision made in the 1990s and not revisited until 2025: the web reserved a status code for payment and then never defined what it meant.

Thirty years of an empty code

402 Payment Required sat in the HTTP status table with the note that it was reserved for future use. The web funded itself with advertising and card rails instead. Micropayments were attempted repeatedly and failed on the same obstacle every time: a card transaction has a floor of roughly thirty cents in fees, and no amount of protocol design makes a three-cent API call viable underneath it.

What changed was not the protocol design. It was that a settlement asset arrived which costs a fraction of a cent to move and settles in seconds.

6 May 2025 — Coinbase gives 402 a body

Coinbase Developer Platform published x402 with Erik Reppel, Nemil Dalal and Dan Kim on the byline. The flow is deliberately small:

  1. The client makes an ordinary GET.
  2. The server answers 402 with its payment requirements — amount, asset, recipient, network.
  3. The client signs a payment authorisation, usually a gasless EIP-3009 USDC transfer on Base.
  4. A facilitator verifies and settles it.
  5. The server returns the resource.

No account, no API key, no session. The client can be a program that came into existence four seconds ago. That is the whole point: x402 is an authorisation protocol for parties that have never met and never will.

It is also why x402 alone is not enough. The protocol does not know who paid, only that payment cleared. Identity is a separate problem, which is why ERC-8004 exists and why the join between a payer address and a named agent is a distinct piece of work. The identity half →

23 September 2025 — Cloudflare, and the move to neutrality

Cloudflare wired 402 into its Agents SDK and MCP hosts, and it and Coinbase said publicly that x402 should sit in a foundation. A payment standard owned by one exchange is a feature; the same standard under neutral governance is a rail. The distinction decides whether a competitor can adopt it.

Cloudflare's pay-per-crawl work is the web2 expression of the same idea: a server that can charge a non-human client per request, without either party holding an account with the other.

April and July 2026 — the Linux Foundation

The Linux Foundation took the project in April 2026. The date, 4/02, is not an accident. The foundation went operational on 14 July 2026 with forty members.

The membership list is the part worth reading closely, because it is a payments list rather than a crypto list: Visa, Mastercard, American Express, Stripe, Adyen, AWS, Google, Circle, Cloudflare, Shopify, Solana Foundation, Ripple, Fiserv. Card networks and processors joining a protocol whose premise is that cards cannot price a three-cent call is the strongest available signal that the sub-dollar machine-to-machine segment is considered real — and that nobody intends to let one company own it.

The Linux Foundation now stewards both A2A and x402, so neither Google nor Coinbase owns the handshake or the payment. The handshake half →

Who is who on this rail

The x402 ecosystem by role. Sellers are not a layer.
RoleWhoWhat they actually do
Author and facilitatorCoinbaseCreated the protocol, runs a facilitator, publishes a directory of paid endpoints.
Specification ownerLinux Foundation x402 projectNeutral governance from 2026. Forty members.
Settlement assetCircleUSDC. Roughly all of the measurable agent ticket volume is denominated in it.
Edge and toolingCloudflare402 in the Agents SDK and MCP hosts; pay-per-crawl for human-facing content.
IndexersPublic x402 explorersCatalog resources and recipients. Their totals disagree with each other by close to an order of magnitude, which we publish rather than resolve.
SellersEveryone with an endpointA 402 wrapper around an API is a seller, not a new layer. Fine, and not a protocol.

What the rail looks like when you measure it

Three things are consistently true of the x402 traffic we can see, and all three are unflattering to the headline numbers:

  • Tickets are tiny. Median payments sit in cents. This is the segment cards structurally cannot serve, which is the entire economic argument for the protocol — and it means dollar volume will always look small next to any DeFi number.
  • The public counters disagree. Two independent x402 indexes report transaction totals that differ by a factor of — right now, because they index different chain sets and different definitions of a transaction. That multiple moves; we publish it live and name the gap instead of picking a winner.
  • A large share of visible traffic is testnet. — of x402 transactions we can see are on Base Sepolia rather than Base mainnet. No public x402 headline separates the two. We do.

The live statistics, with definitions and windows →

The honest limit

x402 does not know who paid. It knows a payment cleared from an address. Everything downstream of that — whether the payer is a named agent, an anonymous script, or fifty wallets funded from one source retrying the same endpoint — is an identity question that the payment protocol correctly does not try to answer.

That gap is the reason this site exists, and it is also the reason our own coverage of the rail is small: we can classify only the payers we have already bound to an identity. We publish that fraction rather than implying our sample is the rail. How a payer becomes an identified actor →

Questions people actually ask

What is x402?
A payment protocol that gives HTTP status code 402 a defined meaning. A server responds 402 with its payment requirements, the client signs a stablecoin authorisation — usually a gasless EIP-3009 USDC transfer on Base — a facilitator verifies and settles it, and the server returns the resource. There is no account and no session, which is what makes it usable by a program that has never interacted with the server before.
Why did HTTP 402 go unused for thirty years?
Because no settlement asset made it worth defining. Card rails carry a fee floor of roughly thirty cents per transaction, so a three-cent API call cannot be priced underneath them at any volume. The web funded itself with advertising and subscriptions instead. Stablecoins changed the arithmetic, not the protocol design.
Who governs x402 now?
The Linux Foundation, which took the project in April 2026 and made it operational on 14 July 2026 with forty member organisations, including Visa, Mastercard, American Express, Stripe, Adyen, AWS, Google, Circle, Cloudflare and Shopify. The same foundation also stewards Google's A2A, so neither Google nor Coinbase owns the handshake or the payment.
Does x402 know which agent made a payment?
No, and deliberately so. x402 proves a payment cleared from an address; it makes no claim about who controls that address. Connecting a paying address to a named agent is a separate problem that ERC-8004 identity and wallet binding exist to solve. A protocol that tried to do both would be adopted by neither side.
How much volume moves over x402?
Less than headline numbers suggest, and the public counters disagree sharply: two independent indexes report transaction totals that differ by a multiple we publish live rather than resolve, because they cover different chain sets and different definitions of a transaction. Tickets are also genuinely small, with medians in cents, and a substantial share of visible activity is on testnet rather than mainnet. We publish both counters and the gap between them rather than choosing one.