Payments · July 29, 2026 · 7 min read
Human buyers find software services through a familiar stack: a search engine, a review aggregator, a pricing page, a trial signup, a procurement approval, a credit card on file. Each step involves a person who can read a landing page, compare alternatives, and make a judgment call. The process is slow, expensive, and deeply human. It also works well enough that it became the default model for every SaaS marketplace ever built.
Autonomous agents cannot do any of that. An agent that needs a real-time market data feed, a voice transcription pipeline, or a language model inference call does not browse G2 Crowd. It needs a way to discover what services exist, what they cost, what they return, and how to pay for them — without a human in the loop. The entire B2B SaaS discovery stack is a human interface problem. Machines need a different solution.
HTTP 402 handles the payment layer. The x402 bazaar protocol handles the discovery layer. Together, they replace the SaaS procurement process with a machine-native flow that can execute in milliseconds.
The core primitive of x402 service discovery is the service manifest: a structured JSON document published at a well-known endpoint that describes a paid service in terms a machine can consume. The manifest is not a marketing page. It does not contain prose descriptions, screenshots, or feature lists. It contains the fields an autonomous agent needs to decide whether to call the service, how to form the request, and how to settle the payment.
A manifest includes the service name and version, the base URL for requests, the supported settlement rails and token addresses, the price per unit and the metering scheme (per-call, per-byte, per-minute), rate limits per buyer wallet, authentication requirements beyond the 402 payment signal, and a JSON schema for the request and response payload. Each field is typed, versioned, and machine-parseable.
The manifest for our market data endpoint, published at agents.ai-rook.com/.well-known/x402, covers 63 endpoints across market data, analysis, media processing, and search. An autonomous agent can retrieve the full manifest in a single request, parse it, evaluate which endpoints match its requirements, and begin making metered requests — all without a human having identified the service in advance.
ai-rook/market-data/v1)A manifest describes one service. A registry aggregates manifests from multiple providers and makes them searchable. The x402 registry is not a storefront — it does not rank providers by reviews, run A/B tests on conversion rates, or take a commission on transactions. It is a structured index: a distributed database of service manifests, queryable by capability, price, settlement rail, and latency profile.
The distinction matters because it determines what the registry can and cannot do. A storefront optimizes for the seller's revenue or the marketplace's take rate. An index optimizes for findability. An autonomous agent that needs a speech-to-text endpoint operating at below $0.005 per minute does not want to be presented with a ranked list of sponsored results. It wants to find the cheapest provider that meets its SLA requirements, read the manifest, and call the endpoint. The registry's job is to make that possible, not to interfere with the selection.
Our registry entry covers the full x402 endpoint manifest at agents.ai-rook.com/.well-known/x402. Providers that publish their manifests to the same registry become discoverable by any autonomous agent that queries the registry. The agent compares manifests programmatically, selects the best fit for its requirements, and initiates the payment flow. No app store. No approval process. No sales call.
The full flow from discovery to settlement runs in four steps and completes in under a second for a well-connected agent.
First, the agent queries the registry for services matching its requirements — for example, any endpoint that provides real-time BTC/USD orderbook data, settles on Base, and prices below $0.01 per call. The registry returns matching manifest IDs and endpoints. Second, the agent reads the manifest for each candidate, validates the JSON schema, confirms the price and metering scheme, and checks the rate limits against its own expected call volume. Third, the agent forms a request and sends it to the selected endpoint. The server returns HTTP 402 with a signed payment intent: the amount, the token, the rail, and a server-side commitment to deliver the requested service. Fourth, the agent's wallet signs and submits the settlement transaction on-chain. The server detects the confirmed transaction and returns 200 with the data. The payment is the authentication. The transaction hash is the receipt.
Steps one through four execute without a human. The agent's wallet holds a balance of USDC on Base. The registry lookup, manifest parsing, payment intent acceptance, and on-chain settlement all happen in the agent's request loop. The only human involvement is the wallet funding — and even that is trending toward programmatic: a liquidity management contract that refills the agent's wallet when the balance drops below a threshold.
Traditional B2B service procurement relies on reputation: the buyer's team has used the vendor before, or a peer recommended them, or a Gartner Magic Quadrant placed them in the upper right. Reputation is a human trust signal. Machines do not have peers or analyst subscriptions. They need a different trust model.
The x402 trust model is structural. An autonomous agent does not trust the service provider because the provider has a good reputation. It trusts the provider because the payment flow is atomic — the agent only pays if the service delivers, and the on-chain settlement transaction is verifiable by anyone. If the provider delivers incorrect data, the agent has an immutable transaction record showing what it paid, what it received, and when. The receipt — combined with the HCS anchor we described in the x402 deployment article — creates an auditable proof trail without requiring trust in the provider's word.
Structural trust also comes from the manifest itself. A manifest with a locked version number, a fixed price, and a published JSON schema is a contractual offer. The provider cannot unilaterally change the price after the agent reads the manifest — the manifest is part of the settlement protocol. If the provider changes the manifest, the agent detects the version bump on the next read and can choose to reject the new terms. This is the machine-equivalent of a contract, expressed in JSON and enforced by the protocol rather than a legal system.
The trading domain is where service discovery for autonomous agents has the most immediate economic value. Professional market data — exchange-grade orderbook feeds, trade tape, funding rates across exchanges — has historically required institutional licensing agreements: annual contracts, procurement reviews, data infrastructure teams, and credit terms. The data is valuable precisely because the access barrier is high.
x402 service discovery changes that access calculus. A trading agent that needs Binance perpetual funding rate data can query the registry, find the relevant manifest, confirm the price, and initiate a metered data session — paying per call, settling on-chain, with no contract and no credit application. The provider sets the price per call. The buyer decides whether the price is worth the data quality. Both parties have an immutable settlement record.
This is not a replacement for institutional data licensing at the high end — a Bloomberg terminal with 30 years of tick history serves a different buyer than a per-call market data endpoint. It is a new tier: lightweight, machine-native, on-demand market data for autonomous trading agents that need real-time signals without the institutional overhead. The x402 bazaar makes that tier economically viable for providers and accessible to buyers.
It is worth being explicit about what the x402 registry is not. It is not a store. It does not hold funds, take a commission, or intermediate the settlement. It does not rank providers by conversion rate or favor providers that pay for placement. It does not make sellers more discoverable by algorithm — it makes them discoverable by specification. The difference matters for the long-term economics: a registry that optimizes for its own take rate will eventually optimize for providers that can pay the take rate, not providers that deliver the best service. A registry that stays infrastructure — visible, neutral, and append-only — creates value for both parties without creating a new intermediary with its own agenda.
The x402 bazaar at agents.ai-rook.com/.well-known/x402 is live and serving the full 63-endpoint manifest. Providers that publish compatible manifests to the same well-known path become part of the same discoverable registry. The protocol is open; the infrastructure is production-grade; the settlement layer is on-chain. Autonomous agents that need services can find them. Providers that have services can be found. That is the entire value proposition, and it is enough.
The full service manifest is available at agents.ai-rook.com/.well-known/x402. Settlement integration documentation is at agents.ai-rook.com.