# Web page reader and scraping API for AI agents: from $0.001 per call ($1 per 1,000), no subscription

> Whizbrew's Web page reader and scraping API costs from $0.001 per call ($1 per 1,000 calls); a typical call is $0.001 ($1 per 1,000). Only a call that returns data is charged: an empty or failed call is free. There is no subscription and no monthly fee. A new account starts with $2 of free credit, with no card.

It reads any public web page as clean text, its contacts, tech, links, forms and screenshots: 41 tools, live at the moment of the call, over one API key or one MCP server (`https://whizbrew.com/mcp`). It is an API, not a scraper you run: no proxies, no browser and no scraping code on your side. The same key reaches 893 tools for other platforms ([all APIs](https://whizbrew.com/api), [all tools](https://whizbrew.com/tools?a=76b5cd3614ed), [pricing](https://whizbrew.com/pricing?a=76b5cd3614ed)).

## Pricing

| Tool | What it returns | Per call that returns data | Per 1,000 calls |
|---|---|---|---|
| `web.snapshots` | Saved copies of a page in the Internet Archive, newest first: capture time, status, size and a link to each copy. Shows how a page changed over time. | $0.001 | $1 |
| `web.snapshot_at` | The saved copy of a page in the Internet Archive closest to a date (the newest one when no date is given). | $0.001 | $1 |
| `web.video_search` | Search videos across the web (YouTube, Rumble, TikTok, Vimeo, news sites): title, URL, site, channel, duration, views and publish date. Views are as the search index last saw them. | $0.002 | $2 |
| `web.reviews` | The ratings and reviews a page publishes in its structured data (schema.org AggregateRating and Review, as JSON-LD or microdata): per rating what it is about, the score, its scale and the review count; per review the stars, title, text, reviewer name, date and publisher. Works on review sites, directories, shops and businesses' own sites that carry this data; a page without it answers with empty lists. The numbers are the page's own statement and are not checked. | $0.001 | $1 |
| `web.meta` | The tags a page describes itself with: title, meta description, canonical, language, robots, Open Graph and Twitter card tags, icons, manifest, feed links and every other named meta tag (site verifications, app ids). Read from the HTML as the server sends it; no script is run. | $0.001 | $1 |
| `web.link_preview` | How a link to a page looks when it is shared: the title, description and image a chat app or social network would show, the site name, favicon and card type, and which tag each one was taken from (Open Graph first, then Twitter tags, then the page itself). | $0.001 | $1 |
| `web.structured_data` | The structured data of a page: every JSON-LD block (parsed, long texts and lists shortened), the microdata items with their properties, the schema.org types found and which of them search engines show rich results for. Invalid JSON-LD blocks are counted. | $0.001 | $1 |
| `web.headings` | The heading outline of a page (h1 to h6 in page order) with the count per level and what is wrong with it: no h1, several h1, skipped levels, empty headings. | $0.001 | $1 |
| `web.links` | Every link of a page: address, link text, internal or external, nofollow / sponsored / ugc, opens a new tab. With totals and the external domains linked most. Each address is listed once. No script is run, so links a script adds are not seen. | $0.001 | $1 |
| `web.images` | The images of a page: address, alt text (null when the attribute is missing), declared width and height, format, lazy loading, responsive sources. With counts of missing and empty alt texts. Set "weigh" to also get the file size of the first 25 images. | $0.001 | $1 |
| `web.hreflang` | The language versions a page declares (hreflang, in its HTML and its Link header) with the checks search engines apply: a self reference, an x-default, valid codes, no code twice, and whether the first 10 other versions link back to this page. Entries declared in a sitemap are not read. | $0.002 | $2 |
| `web.text` | The readable text of a page, one block per line: the part the page marks as its content (main or article), or the page without navigation, header, footer and side columns. With the word count. No script is run, so a page that is built in the browser gives little or no text. | $0.001 | $1 |
| `web.readability` | How hard a page or a text is to read: Flesch reading ease, Flesch-Kincaid grade, Gunning fog, SMOG, ARI and Coleman-Liau, with words per sentence, long sentences and reading time. Give a page URL or the text itself. The formulas are made for English; other languages get the counts and a rough guide. Needs 30 words. | $0.001 | $1 |
| `web.keywords` | The words and two- and three-word phrases a page uses most in its main text, with count and density and whether each is also in the title, the h1, the meta description and the URL. Common words of the page's language are left out. Set "keyword" to check one phrase: count, density and placement. | $0.001 | $1 |
| `web.serp_preview` | A computed preview of a page as a search result: the title as shown (cut at about 580 pixels), the description as shown on a desktop and a phone, the address line, site name, date and the rich result types its structured data allows, with the problems found. An estimate from the page's tags, not a live results page. | $0.001 | $1 |
| `web.language` | The language a page or a text is written in, detected from its script and most common words (21 languages by words, 9 more by script), next to what the page declares (html lang, Content-Language, og:locale) and whether the two agree. Give a page URL or the text itself. | $0.001 | $1 |
| `web.redirects` | The redirect chain of an address, hop by hop (status, target, time), up to 10 hops, with the problems found: temporary redirects, chains, https going back to http, loops, a meta refresh on the final page. For a domain's home page it also checks that http, https, www and non-www all end at one address. | $0.002 | $2 |
| `web.headers` | The response headers of a page with what they mean: server and CDN, compression (encoding, bytes sent and saved), caching rules and CDN cache status, whether HTTP/3 is offered, and the cookies set (names and flags, never values). The HTTP version of the connection and the TLS certificate are not reported. | $0.001 | $1 |
| `web.security_headers` | A grade (A+ to F) for the security headers of a page: https and the http redirect, Strict-Transport-Security, Content-Security-Policy, framing protection, nosniff, Referrer-Policy and Permissions-Policy, each with its value, points and a plain reason. Also notes cookies without Secure or SameSite and headers that name software versions. Our own scale, shown in the answer. | $0.001 | $1 |
| `web.tech` | The technology a website shows it uses, from our own rule list of about 190 products matched on the page's HTML, response headers and first cookies: CMS, site builder, shop system, payments, framework, analytics, tag manager, ad pixels, marketing and email tools, chat, fonts, CDN, hosting, server. Each with the mark that gave it away, plus tag manager and analytics ids. A tool that a tag manager loads later is not found. | $0.001 | $1 |
| `web.consent` | Cookie consent on a page: the consent tool in its HTML (OneTrust, Cookiebot, Usercentrics, Didomi and about 30 more), the consent frameworks it sets up (IAB TCF, GPP, US privacy, Google consent mode), the cookies the first answer sets before any choice (name, lifetime, kind) and the trackers present in the HTML. A banner that only a script loads later is not found; an answer with nothing found is free. | $0.001 | $1 |
| `web.contacts` | The contact details a website publishes: email addresses (links, page text, structured data, also the ones hidden by email protection), phone numbers (phone links and structured data), postal addresses and whether there is a contact form. Read from the page given and the site's contact, about and imprint pages; each result names its page. "contacts" lists every way to reach the site in one typed list, also the social profiles the page links and the page that holds the contact form, for a site that prints no address. Only what the site itself shows. | $0.002 | $2 |
| `web.social_links` | The social profiles and app store pages a website links to (Facebook, Instagram, X, LinkedIn, YouTube, TikTok, Pinterest, GitHub, Threads, Bluesky, Telegram, Discord and about 20 more), each with its handle, from the page's links, rel=me links and structured data. Share buttons and links to single posts are left out. | $0.001 | $1 |
| `web.brand_assets` | A site's brand as its pages show it: name, logo addresses, icons, the sharing image, theme colour, the colours its style sheets use most (brand colours and neutrals apart, with CSS variables named primary, brand or accent) and its fonts (which ones the site loads itself, Google Fonts). Counted from the page and its first 4 style sheets: the most used ones, not a brand guide. | $0.002 | $2 |
| `web.feeds` | The RSS, Atom and JSON feeds of a site: the ones a page declares or links, each opened to check it is a feed, with its title, item count and the date of the newest item. When the page names none, nine usual addresses (/feed, /rss.xml, /atom.xml ...) are tried. Read a feed's posts with newsletter.posts. | $0.002 | $2 |
| `web.page_weight` | What a page makes a browser download, as far as its HTML names it: the HTML (size, compression, inline script and style), then each script, style sheet, image, preload, frame and media file with its size over the wire, totals by type, the largest files and the third-party hosts. Up to 100 files. What scripts load later is not counted, so a browser downloads more. | $0.002 | $2 |
| `web.broken_links` | Checks the links of one page: each different link is asked once (up to 60) and listed when it is broken (not found, server error, a host that does not exist), when the site refuses automated checks (could not tell), or when it redirects. One page a call, not a crawl of the site. | $0.002 | $2 |
| `web.sitemap_stats` | Statistics of a site's sitemap: how many pages it lists, by section (the first part of the path), how fresh they are (newest, oldest, how many changed in the last 7, 30, 90 and 365 days) and whether it lists images, videos, news or language versions. Starts at robots.txt or /sitemap.xml and reads up to 30 sitemap files; a larger site gets a sample and says so. For the pages themselves use domain.sitemap. | $0.002 | $2 |
| `web.app_links` | The apps a website hands its links to: the iOS apps of its apple-app-site-association file (app id, team id, bundle id and the paths each app opens, password autofill, app clips) and the Android apps of its assetlinks.json (package, whether it opens the site's links). Shows which apps belong to a domain. A site with neither file is free. | $0.001 | $1 |
| `web.security_txt` | A site's security.txt (how to report a security problem): contacts, expiry date and whether it has expired, policy, encryption key, acknowledgments, hiring page, with what breaks the standard. A site without the file is free. | $0.001 | $1 |
| `web.humans_txt` | A site's humans.txt (the people and tools behind the site): its sections and entries, and the text. A site without the file is free. | $0.001 | $1 |
| `web.llms_txt` | A site's llms.txt (its guide for AI assistants): title, summary, the sections with their pages (title, address, note), the number of links and whether a long form (llms-full.txt) exists. Shows whether and how a site prepares its content for AI answers. A site without the file is free. | $0.001 | $1 |
| `web.manifest` | A site's web app manifest: app name, start address, display mode, theme and background colour, icons, categories, the store apps it points to, and whether it has the basics a browser needs to offer installing it. Found through the home page or at the usual addresses. A site without one is free. | $0.001 | $1 |
| `web.changes` | What changed on a page between two dates, from the copies the Internet Archive saved nearest to them: title and description before and after, lines of the main text added and removed, headings and link targets added and removed, word counts and how much of the text stayed. Good for watching a competitor's pricing or home page over time. When both dates lead to the same copy the answer says so and is free. | $0.002 | $2 |
| `web.read` | One web address as clean, readable content. An HTML page gives its main content (navigation, banners, cookie layers, footers and side columns left out) as Markdown, plain text or cleaned HTML, with title, author, dates, language, canonical address and word count; tables are kept, links and images on request. A page that only fills in with JavaScript is loaded once in our own headless browser (render: auto) and the answer says so. A PDF gives its text page by page (a scan without a text layer says so: no text recognition). Feeds, plain text, Markdown, CSV, JSON and XML pass through. Long content is cut at max_chars: ask again with offset = next_offset. The site's robots.txt is honoured and the answer says when a page was not read for that reason; a bot check is reported, never solved. | $0.002; with `render` = never: $0.001 | $2 |
| `web.read_many` | Up to 10 web addresses read as clean content in one call, each on its own and all at the same time, the way web.read reads one (HTML pages, PDFs, feeds, text files; a JavaScript page in our headless browser). Each entry is a web.read answer cut at max_chars, or says why that address was not read (error and message): a failed address does not fail the call. Charged when at least one address was read. | $0.01 | $10 |
| `web.search_read` | A web search and the pages of its first results in one call: sources with their content, ready to quote. The search is our search.web (by default the merged list of several engines, or one results page of bing, seznam, yahoo_jp, naver or daum: thin pages, not Google results; the answer names the engines and the limits), then the first results are read like web.read, all at the same time, each cut at max_chars. A result whose page could not be read keeps its title and snippet and says why. | $0.01 | $10 |
| `web.screenshot` | A PNG picture of a web page as our own headless browser draws it: the first screen at the width and height given, or the whole page (full_page, cut at 6,000 pixels of height). The page's scripts run; standard consent banners are removed, media and trackers are not loaded, nothing is clicked and no login or bot check is passed. The image comes back as base64 text. The site's robots.txt is honoured. | $0.005 | $5 |
| `web.affiliate_links` | The affiliate and partner links on a page: each link with its network (Amazon Associates, Awin, CJ, ShareASale, impact.com, Rakuten, Skimlinks, PartnerStack, Rewardful and about 30 more), the partner id where the address carries one (Amazon tag=, ?via=, ?ref=), and whether it is marked rel=sponsored; with totals per network and the disclosure phrase when the page has one. Shows how a publisher or creator earns and which programmes a competitor's reviewers promote. Read from the HTML as sent; cloaked /go/ links and links a script adds are not seen. A page without one is free. | $0.001 | $1 |
| `web.forms` | The forms on a page and the tools behind them: each form's purpose (newsletter signup, lead form, contact, search, login, sign-up, add to cart), where it posts, its visible fields (name, type, required) and its button text, and the email or form tool it belongs to (Mailchimp, Klaviyo, HubSpot, Kit, beehiiv, ActiveCampaign, Brevo, Typeform, Tally, Calendly, Contact Form 7, Gravity Forms and about 30 more), plus forms embedded by script or frame. Shows how a site captures leads and what it asks for. Read from the HTML as sent; a pop-up a script builds later is not seen. A page without a form is free. | $0.001 | $1 |
| `web.oembed` | The embed card of a public post, video, track or photo, by its link, from the platform's own oEmbed service: title, author name and link, type, thumbnail with its size, embed width and height, and the embed HTML. Covers YouTube, Vimeo, TikTok, Spotify, SoundCloud, Reddit and Flickr; any other link is refused. Unlike web.link_preview, which reads the page's own tags, this asks the platform directly, so it also works where the page itself shows a reader a login or consent wall. No counts (views, likes) and no private posts. One read of the platform's public service, no login, no key, no database of ours. A link the platform has no public embed for is free. | $0.001 | $1 |

## Example calls

`web.snapshots` ($0.001 per call): Saved copies of a page in the Internet Archive, newest first: capture time, status, size and a link to each co

```
curl -s -X POST https://whizbrew.com/v1/call/web.snapshots \
  -H "Authorization: Bearer $WHIZBREW_KEY" -H "content-type: application/json" \
  -d '{"limit":3,"url":"example.com"}'
```

`web.snapshot_at` ($0.001 per call): The saved copy of a page in the Internet Archive closest to a date (the newest one when no date is given)

```
curl -s -X POST https://whizbrew.com/v1/call/web.snapshot_at \
  -H "Authorization: Bearer $WHIZBREW_KEY" -H "content-type: application/json" \
  -d '{"date":"2020-01-31","url":"example.com"}'
```

`web.video_search` ($0.002 per call): Search videos across the web (YouTube, Rumble, TikTok, Vimeo, news sites): title, URL, site, channel, duration

```
curl -s -X POST https://whizbrew.com/v1/call/web.video_search \
  -H "Authorization: Bearer $WHIZBREW_KEY" -H "content-type: application/json" \
  -d '{"limit":3,"query":"how to tie a tie"}'
```

Over MCP the same call is `call_tool` with `{"tool": "<name>", "arguments": {…}}`. How often each tool works, measured from real calls: [https://whizbrew.com/status](https://whizbrew.com/status).

## Start

- **An agent on its own**: read [https://whizbrew.com/agents.md?a=76b5cd3614ed](https://whizbrew.com/agents.md?src=body-text&a=76b5cd3614ed). It asks for a free quote (`GET https://whizbrew.com/v1/agent/guide?goal=<the task in plain words>`), then one request (`POST https://whizbrew.com/v1/agent/start?a=76b5cd3614ed`) gives it a key that works at once with the free credit, with no email and no sign-up form. Its human claims the account later.
- **A chat app** (ChatGPT, Claude, Gemini): add the MCP server `https://whizbrew.com/mcp` as a custom connector and sign in. No key to copy.
- **An agent with a shell** (Claude Code, Codex, Cursor): read [https://whizbrew.com/llms.txt?a=76b5cd3614ed](https://whizbrew.com/llms.txt?src=body-text&a=76b5cd3614ed). The agent starts a pairing with one request, its user approves with one click, and the agent holds its own key.
- **Plain HTTP**: create an account at [https://whizbrew.com/app#signup](https://whizbrew.com/app#signup), make a key, then `POST https://whizbrew.com/v1/call/<tool>` with `Authorization: Bearer <key>`.
