readable.

A paid HTTP API that turns any URL into clean, LLM-ready Markdown — boilerplate stripped, headings preserved, real GFM tables, fenced code blocks with detected languages, links and images made absolute. Built for AI agents. Pay per call in USDC on Base with x402; no account, no API key.

Quick start

Any x402 client pays automatically. With the reference client:

npm i @x402/fetch @x402/evm viem

import { wrapFetchWithPayment } from "@x402/fetch";
import { createSigner } from "@x402/evm";

const pay = wrapFetchWithPayment(fetch, createSigner("base", process.env.PRIVATE_KEY));

const res = await pay("https://readable.x.c00l.site/extract", {
  method: "POST",
  headers: { "content-type": "application/json" },
  body: JSON.stringify({ url: "https://en.wikipedia.org/wiki/Markdown" }),
});
const { title, content, wordCount } = await res.json();
console.log(title, wordCount);
console.log(content.slice(0, 500));

Or see the challenge yourself with plain curl — the 402 response carries the payment requirements in the PAYMENT-REQUIRED header:

curl -si https://readable.x.c00l.site/extract \
  -H 'content-type: application/json' \
  -d '{"url":"https://example.com/"}' | head -20

# decode the challenge
curl -si https://readable.x.c00l.site/extract -H 'content-type: application/json' \
  -d '{"url":"https://example.com/"}' \
  | grep -i '^payment-required:' | cut -d' ' -f2 | base64 -d | jq

POST /extract — $0.004

GET /extract?url=… works identically, if a query string suits you better.

Response

{
  "url": "https://example.com/",
  "finalUrl": "https://example.com/",        // after redirects
  "title": "Example Domain",
  "byline": null,                            // author, from JSON-LD / meta / visible byline
  "publishedAt": null,                       // ISO 8601 when parseable
  "siteName": "example.com",
  "lang": "en",
  "excerpt": "This domain is for use in illustrative examples…",
  "content": "# Example Domain\n\nThis domain is for use in…",
  "contentFormat": "markdown",
  "wordCount": 30,
  "readingTimeMinutes": 1,
  "truncated": false,
  "meta": {
    "description": "…", "ogImage": "…", "favicon": "…", "canonical": "…",
    "type": "article", "section": null, "keywords": ["…"],
    "modifiedAt": null, "themeColor": "#fff",
    "contentType": "text/html", "charset": "utf-8"
  },
  "extraction": {                            // how the content block was chosen
    "strategy": "semantic:article", "container": "article",
    "linkDensity": 0.04, "candidateTextChars": 8123, "documentTextChars": 19004,
    "images": 3, "links": 41, "htmlBytes": 214553, "tokens": 6021
  }
}

POST /unfurl — $0.001

Body {"url":"…"}. Stops reading at </head>, so it stays cheap on enormous pages. Returns title, description, siteName, canonical, ogImage, favicon, icons[], type, lang, themeColor, author, publishedAt, twitterCard, oembed, feeds[], keywords[]. Use it to decide whether a URL is worth extracting.

curl -s https://readable.x.c00l.site/unfurl -H 'content-type: application/json' \
  -d '{"url":"https://github.com/cloudflare/workerd"}'

POST /links — $0.002

Body {"url":"…","limit":1000}. Every <a href> on the page, resolved against <base>, with utm_*, fbclid, gclid and friends stripped, deduped by URL + anchor text, and each entry labelled:

• type — link, anchor, file, email, phone
• internal — same registrable site as the page
• nofollow / sponsored / ugc — from rel
• region — nav, header, footer, aside, content, body
• count — how many times that link appears

POST /batch — $0.003 per URL

Body {"urls":[…]}, 1 to 20 URLs, plus any /extract option. The charge is exactly 0.003 × urls.length, so it is fully predictable before you pay. URLs are fetched concurrently; each result carries its own ok flag and, on failure, an error. If every URL fails the response is a 502 and nothing is charged.

curl -s https://readable.x.c00l.site/batch -H 'content-type: application/json' \
  -d '{"urls":["https://example.com/","https://example.org/"],"maxChars":20000}'
# => charged $0.006

How the extraction works

• Streaming parse. One pass of the runtime's native HTMLRewriter. No DOM, no regex over the markup, no headless browser.
• Boilerplate removal. script, style, nav, aside, footer, forms, dialogs and [role=navigation|banner|complementary] go first; then ~250 class/id patterns for ads, cookie banners, share bars, comment threads, related-content rails, newsletter prompts and screen-reader-only furniture.
• Block selection. Prefer <article>, [role=main], <main>. Otherwise score every candidate block on text length discounted by the text sitting inside links — a link-heavy block is navigation, not prose — weighted by tag, class sentiment, and paragraph, heading, comma and code-block counts. Falls back to <body> rather than returning nothing. extraction.strategy tells you which path fired.
• Markdown conversion. h1–h6, emphasis, inline code with adaptive backtick fences, fenced blocks with the language from class="language-x", correctly indented nested lists, blockquotes, <hr>, hard breaks, real GFM pipe tables with alignment and colspan padding, and definition lists. HTML entities are decoded; Markdown metacharacters in text are escaped so the output round-trips.
• Charset. BOM, then Content-Type, then a bounded <meta charset> prescan — UTF-8, Windows-1252, Shift_JIS, GBK, Big5 and EUC-KR pages all come out intact.
• SSRF-safe. Only http(s); no credentials in the URL; loopback, private, CGNAT, link-local (169.254.0.0/16, i.e. cloud metadata), multicast and reserved IPv4/IPv6 refused; .local, .internal, single-label and numeric hosts refused; sensitive ports blocked. Re-checked on every redirect hop, not just the URL you sent.
• Limits. 10 s upstream timeout, 5 MB body cap, 6 redirect hops. JavaScript-rendered SPAs degrade gracefully to whatever is in the served HTML — we do not run a browser, and we tell you so via a low documentTextChars.

Errors

Always {"error":{"code":"…","message":"…"}}.

Every one of these is a non-2xx, which means settlement is skipped and the call is free.

Payment

x402 v2USDCBase mainnet Send a request; get 402 with a PAYMENT-REQUIRED challenge; sign an EIP-3009 authorisation; resend with PAYMENT-SIGNATURE. Settlement is confirmed before the body is delivered and reported in PAYMENT-RESPONSE. Any x402-capable client — @x402/fetch, an MCP payment proxy, an agent framework — handles this for you.