How do I read any web page as clean text?

Q: 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

Web Extraction APIs

One URL as clean markdown

Say this to your agent

>read <url> 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

What the agent runsdone

context.dev /web/scrape/markdown$0.0009

octen /extract$0.001 / result

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

firecrawl /scrape · not run$0.0002

api.strale.io /x402/url-to-markdown$0.0594

What came backfour real pages · 2026-09-24

Same docs page, two providerscharacters returned

$0.000915,312

$0.05949,948

the cheap one15,312 characters with all three code samples intact, plus the page metadata

the 66x one9,948 characters and every code fence stripped to prose

what that meanson a documentation page, the expensive extractor lost exactly the part you came for

count blocks, not characterson 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 bytesthe endpoint's own content length is counted in utf-8 bytes rather than characters, so it will not match a character count either

the tieon a news article the cheap one and a near-free one returned the same text, neither blocked

$0.0009per page

Search, then read the top sources

Say this to your agent

>search for <question> with context.dev on monid and read the top ten results in the same call, turning the markdown option on explicitly. without it the response carries ten results and not one character of page text, and it bills the same. tell me which came back empty so i know what i paid for

What the agent runsdone

context.dev /web/search$0.00009 / result

tinyfish /search · not run$0.00

litescrape /google/search · not run$0.00015

exa /search · not run$0.002

What came backfour real pages · 2026-09-24

the flowten results found and their pages read in one call, from 11 characters to 50,718

the noisetwo of the ten were empty in practice: an eleven-character stub and a login wall

billed anywayboth empty results were charged as normal hits, so about a fifth of the spend bought nothing

off by default, and it bills anywaywithout the markdown option the ten results came back with a null body and a not-requested code, zero characters of text, at the same $0.0009

the surprisereading the pages was not metered separately from the search, despite the catalog note

$0.0009ten results, searched and read

Crawl a section of a site

Say this to your agent

>crawl <url> with context.dev on monid, five pages maximum, AND pass a url pattern restricting it to that path. without one it leaves the section on the first hop and still reports every page as a success. then read back the urls it actually fetched

What the agent runsdone

context.dev /web/crawl$0.0009 / result

context.dev /web/scrape/sitemap · not run$0.0009

firecrawl /crawl · not run$0.001 / result

ahrefs /site-explorer/crawled-pages · not run$0.018 / result

What came backfour real pages · 2026-09-24

the resultfive of five pages fetched, none failed, between 4,530 and 15,823 characters each

it leaves on the first hopon a second site a naive five-page crawl fetched the seed and then four pages outside the path, none of them in the section asked for, and reported five succeeded and none failed

the parameter that fixes ita url pattern keeps all five inside the prefix; it is not in any of these sentences by default and it is fiddly to quote on a shell, so pass it from a file

the driftthe docs domain had migrated, so one hop out the crawl left the section entirely

what to docap the pages, then read back the urls before trusting the set

the cheaper stepif you only need the list of urls, the sitemap call is one flat $0.0009

$0.0045five pages

Turn a page into structured fields

Say this to your agent

>extract <schema> from <url> with context.dev on monid with maxPages set to 1, and do NOT turn fact checking on: with it off an absent field came back empty, and with it on the same field came back as a string and a missing price came back as zero. check the cache age before you trust anything, and treat any zero as unverified

What the agent runsdone

context.dev /web/extract$0.009

mrscraper /scrape/extract$0.001 / result

api.strale.io /x402/web-extract · not run$0.1782

What came backfour real pages · 2026-09-24

Same pricing page, two approachesrows returned

schema $0.00917 clean

prompt $0.03855 messy

fact checking does not do what it saysasked for two fields a page genuinely does not contain, it returned a five-character string where a phone number should be and a zero where a free-trial count should be, both of which read as real values downstream

and a zero price is the dangerous oneit reported a monthly price of 0 for an enterprise plan whose own page says the price is custom; re-running with the cache disabled produced the identical invention, so it is not a stale answer

off is safer than onwith the flag omitted the same absent field came back as an empty string, which is the honest answer; turning it on is what produced the fabrications

the schema way17 rows, one of them left blank on purpose because the page did not support a value

the prompt way55 rows with invented identifiers and duplicates no field could tell apart

the hidden pricethat one billed $0.038 against a $0.001 catalog rate, because browser rendering was on

it crawls, it does not read one pagethis endpoint defaults to five pages and follows links, so a second run's analysed urls included two product pages nobody asked for and the field list filled with rows from them; set the page count to one if you mean one page

the cachethe clean answer came from a cache almost six days old, with no warning a normal caller would see

$0.009flat, per page

A brand's identity from its domain

Say this to your agent

>get the brand profile for <domain> with context.dev on monid: name, description, colours, logos and socials. tell me which fields came back empty

What the agent runsdone

context.dev /brand/retrieve$0.009

hunterio /companies/find · not run$0.004784

contactout /v1/domain/enrich · not run$0.02 / result

ploid /socials · not run$0.10 / result

What came backfour real pages · 2026-09-24

what came backthe title, description, slogan, four brand colours and two logo variants with their own palettes

and morea head office country, three social profiles and a headcount both as a range and an exact figure

it can invent a contacton a second domain the email field came back as a plausible address at that company's domain, assembled from a demo persona the company publishes with a different domain; treat any contact from here as unverified

and it cannot be read freshthis endpoint's cache defaults to three months and clamps to a one-day floor, so there is no way to ask it for today

the gapsthe email came back as an empty string and three link fields were null, though all three pages exist

the usethis is the cheapest way to skin a page or a deck in someone else's brand without asking them for a kit

$0.009per domain

Web Extraction

Try it with your agents

How-to guide

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