How do I read any web page as clean text?

free route first, escalate on a thin answer read as markdown with context.dev on monid and keep the code blocks, then count the fenced blocks against the page and tell me the number. do not judge it by the character count: link syntax and page chrome are included by default and can make the output larger than the visible text the cheap one: 15,312 characters with all three code samples intact, plus the page metadata the 66x one: 9,948 characters and every code fence stripped to prose what that means: on a documentation page, the expensive extractor lost exactly the part you came for count blocks, not characters: on a second page all 32 code blocks came back and twenty sampled ones were verbatim, while the character count was 74% larger than the visible text because link markup and chrome are on by default and made up two fifths of the output and the length field is bytes: the endpoint's own content length is counted in utf-8 bytes rather than characters, so it will not match a character count either the tie: on a news article the cheap one and a near-free one returned the same text, neither blocked

Browserbase APIs

Q: How do I read a page that blocks a plain fetch?

a real browser session, and what it costs per step read with tinyfish on monid and give me the items on it. it is a page where nothing is in the html until javascript runs, so tell me how many items came back with real values in them. do not judge it by the byte count or by the status: judge it by whether the fields are populated the comparison: on one page the $0.011 route and plain curl returned the same 5,806 bytes and a string equality check said true; on another the bytes differed by 367 and nothing had rendered either way what that means: the script that builds the page was still sitting there unexecuted, so an agent that strips tags and reads text gets nothing back the free one: returned all ten items as 1,080 bytes of clean markdown, which is the whole job, for $0.0000 do not diff the bytes: that 367-byte gap was a CDN performance beacon injected because the request looked like a browser, so on any fronted site the byte check reports a difference that means nothing the rule: count the populated fields. Both routes answered HTTP 200 on both pages, and on two different targets the paid one rendered 0 of 12 and 0 of 10

Read a page that JavaScript builds

Say this to your agent

>read <url> with tinyfish on monid and give me the items on it. it is a page where nothing is in the html until javascript runs, so tell me how many items came back with real values in them. do not judge it by the byte count or by the status: judge it by whether the fields are populated

What the agent runsdone

tinyfish /fetch$0.00

context.dev /web/extract$0.009

context.dev /web/scrape/markdown$0.0009

x402.browserbase.com /fetch$0.011

What came backone JS-built page · 2026-09-25

Same url, same minute, four routesitems actually rendered, of ten

tinyfish · $0.0010

context.dev extract · $0.00910, typed

plain curl · $0.000

browserbase · $0.0110

the comparisonon one page the $0.011 route and plain curl returned the same 5,806 bytes and a string equality check said true; on another the bytes differed by 367 and nothing had rendered either way

what that meansthe script that builds the page was still sitting there unexecuted, so an agent that strips tags and reads text gets nothing back

the free onereturned all ten items as 1,080 bytes of clean markdown, which is the whole job, for $0.0000

do not diff the bytesthat 367-byte gap was a CDN performance beacon injected because the request looked like a browser, so on any fronted site the byte check reports a difference that means nothing

the rulecount the populated fields. Both routes answered HTTP 200 on both pages, and on two different targets the paid one rendered 0 of 12 and 0 of 10

$0.00ten items, rendered

When the free read comes back empty

Say this to your agent

>read <url> with tinyfish on monid first, then check whether the fields i asked for actually came back with values. escalate to mrscraper with browserRendering and super set to true whenever they did not, whether that is a 403, a challenge page, or a 200 whose values are still placeholders. do not send every url through the stealth route: escalate on an empty answer, not on the batch

What the agent runsdone

tinyfish /fetch$0.00

mrscraper /scrape/htmlfrom $0.001 / result

context.dev /web/scrape/markdown$0.0009

x402.browserbase.com /fetch$0.011

What came backone JS-built page · 2026-09-25

A hundred pages, ten of them walledUS dollars

free first, escalate on a thin answer$0.12

everything through stealth$1.20

the honest failureon one walled site the free route returned a 403 inside an errors array with no results and billed $0.0000, which is exactly the signal an escalation needs

the failure with no signalon another it walked through the wall and answered 200 with an empty errors array, a real page title and a real result count, and one listing where there were sixty-two prices, the rest still reading Loading

so check the yielda status code cannot tell those two apart; only counting the values you asked for can, which is why the sentence escalates on an empty answer rather than on a 403

what got throughthe stealth route returned the real page both times, 1.2 MB on one and 62 prices against 9 listings on the other

what it really costsits listed $0.001 per result is a token unit, not an estimate: two runs billed $0.012 and $0.034, so budget an order of magnitude above the sticker

$0.012only on the ones that block

Take a picture of the page

Say this to your agent

>screenshot <url> with context.dev on monid at full page. it returns a hosted image url rather than the bytes, so fetch that url separately if you need the file itself

What the agent runsdone

context.dev /web/screenshot$0.0045

mrscraper /scrape/screenshot · not run$0.001 / result

tinyfish /fetch$0.00

What came backone JS-built page · 2026-09-25

what came backa 1920 by 1698 full-page PNG with the JavaScript-built cards fully drawn, which is independent proof that a real browser ran

what you get in the response329 bytes of JSON holding a url on a third-party CDN, not the image; the file itself was 163,588 bytes on a second, unbilled fetch

the retentionno stated lifetime on that hosted url, so download it in the same run if you need to keep it

the aisle it is not inthe browser aisle has no screenshot endpoint at all; this one lives next door

$0.0045one full-page capture

Typed values, or a regex over free text

Say this to your agent

>extract <fields> from <url> with context.dev on monid, passing a json schema and factCheck true. use this when you want typed values and a stable shape rather than text you have to pattern-match; if a four-line regex over the free markdown already gets you the fields, take the free one

What the agent runsdone

context.dev /web/extract$0.009

tinyfish /fetch$0.00

context.dev /web/scrape/markdown$0.0009

What came backone JS-built page · 2026-09-25

what it gives youevery item in exactly the requested shape with the numbers as numbers, so a price is 52 rather than the string $52.00, and the shape does not move when the page's layout does

when the free route is enoughon one page markdown flattening dropped every tag array; on another it kept all twelve items and both fields, and four lines of adjacency matching recovered them exactly. Check before you pay

the honest price$0.009 a call whatever the schema. The thing you are buying is not completeness, it is not having to write and keep a pattern per site

the grounding flagfactCheck true forces every value back to text that is actually on the page, which is the difference between extraction and invention

$0.009per call, any schema size

Search the live web and read the results

Say this to your agent

>search the web for <query> with tinyfish on monid and give me the results with their snippets, then fetch the ones worth reading in one call rather than one call each

What the agent runsdone

tinyfish /search$0.00

x402.browserbase.com /search$0.011

tinyfish /fetch$0.00

What came backone JS-built page · 2026-09-25

Same query, two search endpointsresults carrying a snippet, of ten

tinyfish · $0.0010

browserbase · $0.0110

the differenceone returned a position, a site name, a snippet, a title and a url on every row; the other returned an id, a url and a title, with two titles empty

why the snippet decides itwithout one you have to fetch all ten urls to know which is worth reading, and the second half of the job is reading them

what that coststhe paid search plus ten paid fetches is $0.121 a query; the free search plus a free multi-url fetch is $0.0000

the batchingthe free fetch takes up to ten urls in a single call, so a hundred pages is ten calls, still free

$0.00ten results with snippets

Click, type and log in

Say this to your agent

>create a browserbase session with monid and connect to the connectUrl ONCE, with playwright's connectOverCDP, and do all your work in that one attach. do not test the socket first and do not close and reopen: the session dies on the first detach and the paid time left on it is forfeit

What the agent runsdone

x402.browserbase.com /browser/session/create$0.011

tinyfish /fetch$0.00

mrscraper /scrape/htmlfrom $0.001 / result

What came backone JS-built page · 2026-09-25

it is a real browsera WebSocket handshake answers HTTP 101 and Playwright's connectOverCDP drives it, twelve items off a JavaScript page on the first attach

what Monid sells youthe browser, not the driving: the session comes with a live view url and an auth token, and the steps are yours to write

one attach, not five minutesafter a clean close, reconnecting 4 minutes 41 seconds before the stated expiry answered 410 Gone, session not running. Even a bare handshake to check the socket burns it. Connect once and stay

it is not a way past a walla browser you drive yourself got the same 403 on a protected site that everything else did; a real browser and a stealth route are different products

$0.011five paid minutes

Browserbase

How-to guides

Set up Monid.

Pick what you need. Say it in one sentence.

One URL as clean markdown

Search, then read the top sources

Crawl a section of a site

Turn a page into structured fields

A brand's identity from its domain

Read more

Extract page content for RAG

When the site will not let you in

Bing Search API retired: the alternatives

Set up Monid.

Pick what you need. Say it in one sentence.

Read a page that JavaScript builds

When the free read comes back empty

Take a picture of the page

Typed values, or a regex over free text

Search the live web and read the results

Click, type and log in

Read more

When the site will not let you in

Extract page content for RAG

Bing Search API retired: the alternatives