Web scraping API
Web scraping for AI agents, paid per page
Your agent sends a URL and gets the page as clean Markdown: title, meta description and the main content without menus, ads and footers. $0.0015 per page, up to 20 pages in one call for $0.03. No sign-up, no API key and no monthly plan: the agent pays each call itself with USDC via x402.
Try it now, free, without a wallet:
curl -X POST https://agentsvc.io/api/v1/try/web-scrape -H "Content-Type: application/json" -d '{"url":"https://example.com"}'Why pay per call
Most scraping APIs are built for a human who signs up, stores a key and picks a monthly plan. An agent cannot do that alone, and a project that reads 300 pages pays for a whole month.
| agentsvc.io | Typical scraping API | |
|---|---|---|
| Sign-up and API key | none | required |
| Minimum spend | one call, $0.0015 | monthly plan or prepaid credits |
| Who pays | the agent, from its own wallet | the account owner |
| Failed pages | not charged | depends on the plan |
| Output | Markdown made for LLMs | often raw HTML |
Need a key anyway, for a script or a team? A credit token works like one, from $5 with 10% bonus.
Endpoints
| Service | Price | Use it for |
|---|---|---|
| web-scrape | $0.0015 | one URL to Markdown; render: true for JavaScript apps |
| web-read-batch | $0.005 | 2 to 5 URLs in one call |
| web-scrape-batch | $0.03 | 6 to 20 URLs, one payment, fetched in parallel |
| site-map | $0.0025 | list the URLs of a site first |
| site-crawl | $0.005 | follow links from a start URL, up to 20 pages |
| search-read | $0.005 | a question but no URL: search, then read the top results |
Each answer of web-scrape has title, description, markdown, word_count, truncated and the final URL after redirects. max_chars goes up to 60,000, include_links adds up to 100 absolute links.
Zero CLI
If your coding agent already uses the Zero CLI, it pays the 402 for you:
zero fetch https://agentsvc.io/api/v1/proxy/web-scrape -d '{"url":"https://example.com/blog/post"}'Claude, Cursor and other MCP clients
Add https://agentsvc.io/mcp/credit as a custom connector. You sign in with an email address and get a small starting credit, then web-scrape is one tool among the others. Agents that pay with their own wallet use https://agentsvc.io/mcp with x402 in the request. Details in the docs.
LangChain, AI SDK, CrewAI
pip install "agentsvc[langchain]" from agentsvc import Agentsvc from agentsvc.langchain import agentsvc_tools svc = Agentsvc(credit_token="asvc_cr_...") # or Agentsvc(private_key="0x...", network="eip155:8453") tools = agentsvc_tools(svc, ["web-scrape", "web-scrape-batch", "search-read"])
npm install agentsvc ai
import { createAgentsvc } from 'agentsvc';
import { agentsvcTools } from 'agentsvc/ai-sdk';
const svc = createAgentsvc({ creditToken: process.env.AGENTSVC_CREDIT_TOKEN });
const tools = await agentsvcTools(svc, ['web-scrape', 'web-scrape-batch']);Full examples: LangChain, Vercel AI SDK, Coinbase AgentKit. Coding agents can learn it as a skill: npx skills add agentsvc/skills.
Plain HTTP with x402
npm i @x402/fetch @x402/evm viem
import { wrapFetchWithPayment, x402Client } from "@x402/fetch";
import { registerExactEvmScheme } from "@x402/evm/exact/client";
import { privateKeyToAccount } from "viem/accounts";
const client = new x402Client();
registerExactEvmScheme(client, { signer: privateKeyToAccount(process.env.EVM_PRIVATE_KEY) });
const payFetch = wrapFetchWithPayment(fetch, client);
const res = await payFetch("https://agentsvc.io/api/v1/proxy/web-scrape-batch", {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify({ urls: ["https://example.com/a", "https://example.com/b"], max_chars: 8000 }),
});
console.log(await res.json());Python, curl and the step-by-step guide: read web pages as Markdown. Current uptime and response times: /stats.
FAQ
Do I need an account or an API key?
No. The first request returns HTTP 402 with the price, the x402 client signs a USDC payment and sends the request again. If you prefer a key, buy a credit token at /credits and send it as Authorization: Bearer.
Am I charged for pages that fail?
No. Invalid input and server errors are never charged. web-scrape-batch is charged only if at least one of its pages was read.
What about JavaScript apps?
Plain pages are fetched directly and returned as Markdown. If a single-page app comes back nearly empty, send render: true: a real browser loads the page, runs its JavaScript and returns the text.
Can it read PDFs?
web-scrape reads HTML, Markdown and plain text (llms.txt and README.md come back unchanged). For a PDF URL use pdf-extract ($0.0055).
Which networks can pay?
USDC on Base, Solana, Polygon or Arbitrum. The paying wallet needs no gas, it only signs a transfer authorization per call.