132 Pay-Per-Call APIs: AI Agents Pay with USDC via x402

AI Agents often need small facts while completing tasks: a token balance, a company's financial report, cleanly formatted web page text, a weather alert. The problem is that the vast majority of APIs require agents to first register an account, store keys, and purchase a plan. For scenarios that only need a single call, this process is both cumbersome and uneconomical.

SiteCheck offers another approach. It turns all 132 tools into a single HTTP endpoint, with per-call prices ranging from $0.001 to $0.01. Agents call it directly, receive an HTTP 402 response with pricing, complete payment in USDC on Base, Solana, or Arc via the x402 protocol, and then get the result. No account, no API key, no subscription. If the upstream data source call fails, the call is not charged.

rss source: 132 pay-per-call APIs for AI agents, paid in USDC with x402

How the x402 Protocol Lets Agents Pay Autonomously

The core of x402 is turning "payment" into a standard HTTP interaction. Traditional API authorization logic lives in account systems and keys, while x402 writes the price directly into the HTTP 402 Payment Required status code. After reading this response, the agent or client completes a USDC transfer on-chain, then replays the same request to get the data.

This design is especially friendly to agents: it does not require humans to preconfigure billing information, nor does it require understanding complex subscription tiers. As long as there is USDC in the wallet, an agent can purchase data on demand. The payment network supports Base, Solana, and Arc, covering the current mainstream low-cost on-chain settlement environments.

HTTP 402 Payment Required is a status code in the HTTP protocol that has long been in a "reserved" state. It has existed since the HTTP/1.1 specification in 1997, but for decades it has hardly been truly used by mainstream web services, because there was no widely accepted machine-readable payment method. The x402 protocol was designed to fill this gap: it specifies that when a server returns 402, it simultaneously carries structured information in the response headers or body, including price, currency, receiving address, chain ID, and so on. The client then completes the on-chain transfer accordingly and sends back the transaction hash through a specific request header. After verification, the server returns the originally requested resource. This design decouples payment from "account state" and turns it into a stateless, replayable request, so it naturally fits the calling pattern of AI Agents—an agent only needs to hold one wallet private key to access any service supporting x402, without establishing a separate account relationship for each vendor.

What Data the 132 Tools Cover

SiteCheck's tool list is divided into five categories by domain, basically centered on the common "look this up" needs of agents.

On-Chain and Crypto Data

This includes balances, transactions, transfers, logs, contract calls, and NFT reads on EVM and Solana, as well as verified contract source code, calldata decoding, and gas fees (including L1 data fees on Base and Optimism). There are also Bitcoin, ENS, Jupiter quotes, Hyperliquid, Polymarket, DefiLlama TVL, yields, fees, and DEX trading volume.

Business and Public Data

This line leans toward traditional finance and compliance scenarios: SEC filings, XBRL financial data, full-text search, 8-K events, and insider transactions; company registry information for the UK, Norway, Finland, Czechia, and Poland; EU VAT, IBAN, LEI; FDA labels and adverse events; clinical trials; PubMed; Grants.gov; EU tenders; congressional bills; and interest rate data from the Federal Reserve, European Central Bank, and Eurostat.

Web Processing and Reference Information

Web tools handle converting pages to Markdown, PDF to text, extracting metadata, checking broken links and redirects, reading security headers, DNS, OCR, tech stacks, logos, certificates, trackers, and llms.txt. Reference tools cover weather and NWS alerts, geocoding, earthquakes, time zones, holidays, IP attribution, software package vulnerabilities, GitHub, Stack Overflow, and Wikidata.

llms.txt is a convention file that has emerged in the past two years, usually placed in a website's root directory, providing large language models and AI Agents with a structured index and usage instructions for the site's content in Markdown format. Its role is similar to robots.txt for crawlers, but its target readers are models rather than search engines: authors can list important page links, documentation entry points, API descriptions, and content licensing terms within it, helping agents skip navigation bars and ads and directly obtain high-quality context. SiteCheck makes reading llms.txt into a tool, meaning an agent can first fetch this "readme file" before crawling a site, thereby determining whether the site welcomes crawling and which pages are most valuable, reducing ineffective requests and parsing costs.

Text AI Capabilities

There is also a set of lightweight text AI tools for schema-based extraction, summarization, classification, entity recognition, redaction, content moderation, and image description. These capabilities are suitable for pre-processing or post-processing in agent workflows, without having to separately integrate another large model API.

How to Try It

Any tool returns a price when requested without payment. For example:

curl -i "https://api.sitecheck-api.workers.dev/api/crypto/price?symbols=BTC,ETH"

An x402 client reads this 402 response, completes payment, and repeats the request. The complete tool list and input/output schemas are at /openapi.json, and the same batch of tools is also available as a remote MCP server:

https://api.sitecheck-api.workers.dev/mcp

For agent frameworks already using MCP, this means lower integration cost, with no need to write a separate adapter layer for each data source.

A remote MCP server refers to a Model Context Protocol service endpoint exposed over HTTP (usually via SSE or Streamable HTTP transport). MCP was proposed by Anthropic at the end of 2024 to allow AI applications to connect to external tools and data sources through a unified protocol. It initially mainly used local stdio processes, and later expanded to a remote mode. For agent frameworks, the benefit of integrating a remote MCP is that there is no need to write separate adapter code for each data source or manage local processes. As long as the framework supports MCP, entering the server address into the configuration lets it see all tools and their input/output schemas. SiteCheck provides both OpenAPI and MCP descriptions: the former targets general HTTP clients and code generators, while the latter targets agent runtimes that already have built-in MCP support. Together they cover the current mainstream integration paths.

Accuracy Verification and Limitations

According to the official statement, this week every tool was run against inputs with known answers, and units, decimal places, dates, and missing fields were compared against official sources. Problematic tools were fixed, while those that could not reliably access upstream sources in the Cloudflare Workers environment were removed.

This "pay-per-call, no charge on failure" model lowers the trial-and-error cost for agents calling external data, but its reliability still depends on upstream data sources and the stability of on-chain payments. For scenarios requiring high-frequency, large-batch calls, the per-call cost of $0.001 to $0.01 also needs to be included in budget evaluation.

One-line Summary

SiteCheck uses x402 to turn 132 data tools into pay-per-call HTTP endpoints, letting agents call them autonomously with USDC.

Paragraph Summary

SiteCheck unifies 132 data query tools into a single HTTP endpoint, with per-call prices ranging from $0.001 to $0.01 and no need to register an account, use an API key, or buy a subscription plan. Payment relies on the x402 protocol: the client first receives an HTTP 402 response and pricing information, completes a USDC transfer on Base, Solana, or Arc, and then replays the request to obtain data. If the upstream data source fails, there is no charge. The tools cover four major areas: on-chain and crypto data, business and public data, web processing and reference information, and text AI capabilities. They are exposed through both OpenAPI and a remote MCP server, making low-cost integration easy for agent frameworks already using MCP. The official statement says every tool has undergone accuracy verification, but the reliability of this model still depends on upstream data sources and the stability of on-chain payments, and high-frequency call scenarios need to evaluate the per-call cost.