Web data API + MCP server

Cheap scraping at scale for AI agents.

Your agent pays once to learn a site. Every page after that is extracted free, as JSON, with the selector behind every value.

Get a free API key Connect to Claude Code

25 free credits · no card · read and map are always free

Try it: no sign-up

Pre-warmed examples, always free

Read your own page (free): Markdown, no sign-up

0 tokens

on every repeat

Bench: npm run bench

~200 ms

warm extraction

Bench: npm run bench

1 credit

($0.01) to learn a site

Pricing

$0.50

per 1k residential pages, only on success

Pricing

Measured live with npm run bench against real sites; prices at $0.01/credit.

When a site blocks cloud servers, Scraipe reroutes under the same honest name and charges only if the page comes back.

How it gets cheap

Pay once to learn a site. Repeats cost $0.

01
Learn
Learned layout · compiled-fresh
1 credit5–15 s · once per site layout
02
Reuse
Reused extractor · compiled
Free~200 ms · every page after
03
Repair
Repaired extractor · healed
1 creditonly when the layout changes

Live bench, quotes.toscrape: first call 3,938 ms and 1,709 tokens; every later call 224 ms and 0 tokens.

Cost per 1,000 pages vs. how often you revisit

Scraipe falls from $1.00 at 10 repeats to $0.01 at 1,000. List prices stay flat.

Flat lines are list prices as of 2026-09-23: Firecrawl pricing (opens in a new tab) (JSON, 5 credits a page at $99 per 100k) and Bright Data Web Unlocker pricing (opens in a new tab) (HTML only). Scraipe at $0.01 per credit, no residential pages.

Where others are cheaper: below about 2 repeats (Firecrawl) and 7 repeats (Bright Data), marked with dots. Scraipe is built for pages you visit more than once.

Networks

When sites block the cloud

Blocked from AWS? We find the cheapest network that works.

Most blocks are about where a request comes from, not who sent it. Scraipe learns, for each site, the cheapest network that works, and remembers it. Every network sends the same name and follows the same robots.txt. We don't solve CAPTCHAs or disguise ourselves. When a site says no, you get the vendor, the networks we tried and what would work instead, and you're not charged.

No CAPTCHA solving, no stealth fingerprinting, no IP rotation to escape a block or a rate limit. When a site blocks cloud servers wholesale, or bans our servers outright, Scraipe may fetch through a residential network, always under the same honest name and always obeying robots.txt.

Run a site? See how to identify or block us

Rung 1: Our servers

free

Every request starts here.

if the site blocks cloud servers, or bans ours

Rung 2: Residential

0.05 credit, only if it works

After your first purchase. A browser render costs 0.20 credit.

if a human check, a rate limit or robots.txt says no

Rung 3: Refuse with a reason

free

The vendor, the networks tried, and what would work.

Same User-Agent and the same robots.txt on every rung. Blocks are never billed.

Receipts, not guesses

Every value comes back with the selector that produced it.

Every value comes from a real DOM node.
The extractor is data, not code.
Same page, same output.

price · Row 1

Selector: article.product_pod p.price_color
Raw text: £51.77
Matches: 20
Index: 0
Transforms: trim → toNumber

Hover or focus a value in the response to see its selector.

POST /v1/scrape · books.toscrape.com

{
  "ok": true,
  "data": [
    { "title": , "price":  },
    { "title": , "price":  },
    … 18 more
  ],
  "provenance": [
    { "price": { "selector": "article.product_pod p.price_color",
      "found": 20, "index": 0, "raw": "£51.77" } },
    … 19 more
  ],  "meta": { "tier": "compiled", "latencyMs": 7, "tokens": 0,
            "creditsCharged": 0, "egress": { "rung": "none" } }
}

Use cases

Price & stock monitoring

A quiet page costs $0 per check.

Watch run · page unchanged (304)Free

RAG & docs ingestion

Read and map are free.

read · MarkdownFree

map_site · URL listFree

Research agents

1 credit per new site layout, then free.

New site layout1 credit

Every page afterFree

Connect your agent in one line

Hosted MCP with key auth. Responses are capped at 8k tokens and return a cursor.

claude mcp add --transport http scraipe https://scraipe-fc66a552921e.herokuapp.com/mcp \
  --header "Authorization: Bearer $SCRAIPE_API_KEY"

Terminal · Add --scope user to use it in every project.

Get a key to fill this in

Endpoint: https://scraipe-fc66a552921e.herokuapp.com/mcp
Auth: Authorization: Bearer sk_scr_…
Your agent sees: tier: compiled · credits: 0 · egress: direct/http

MCP tools and what they cost

API docs

MCP tools and what each one costs
Tool	Cost
read	free on our servers
map_site	free
scrape	1 credit when it learns a layout
crawl	per page, as scrape
watch	free while the page is unchanged
compile	1 credit
batch	per page, as scrape
egress_quote	free
list_extractors	free

Pricing

You pay when Scraipe has to think, or when a page needs a residential network.

Cache, 304 and reused extractorsFree

Learning or repairing a layout1 credit

Residential pagesfrom 0.05 credit

Blocks and failuresFree

Residential is charged only on success. 1 credit = $0.01.

See pricing and the calculator

How the prices compare

Per 1,000 requests, list prices as of 2026-09-23, Scraipe at $0.01 per credit.

Scraipe against Firecrawl, ScrapingBee and Bright Data, per 1,000 requests, as of 2026-09-23
Request	Scraipe	Firecrawl	ScrapingBee	Bright Data
Repeat request, page unchanged	Free	$0.60–0.99	$0.075–2.48	$1.00–1.50
Structured JSON	$0.01 once per site and schema, then free	JSON +4 credits every page: $3.00–4.95	AI extraction +5 credits every page: $0.37–0.50	Markdown or HTML only
Blocked request	Free	A returned 403/404 page costs 1 credit	Free (200 and 404 billed)	Free (billed if custom headers or cookies are sent)

As of 2026-09-23 · sources: Firecrawl (opens in a new tab), ScrapingBee (opens in a new tab), Bright Data (opens in a new tab)

Full comparison

Questions

Short answers, each with a link to the docs.

Something else? support@scraipe.co

Is it really free on repeats?

Yes. Cache hits, unchanged pages (304) and reused extractors cost 0 credits, and they keep working at a zero balance. You pay 1 credit only when Scraipe learns or repairs a layout. How it works

What happens when a site redesigns?

The extractor notices its fields no longer match, and Scraipe repairs it on the next request for 1 credit. Repairs are capped at 3 per layout per day, so a site that keeps changing can't drain your balance. How it notices

Do you get around bot protection?

No. We don't solve CAPTCHAs, forge fingerprints or rotate IPs to escape a block or a rate limit. When a site blocks cloud servers wholesale, Scraipe may fetch through a residential network under the same honest name, obeying robots.txt, and charges only if the page comes back. For site owners

What does a credit cost?

One credit is $0.01. Learning a layout costs 1 credit; a residential page 0.05 (0.15 over 256 KB) and a residential browser render 0.20, only on success. Pricing

Can my agent run up a bill?

No. Credits are prepaid, with no subscription or auto-recharge, so spending stops at zero. Blocked and failed requests are never charged, and POST /v1/quote prices a request before you run it. API reference

Your agent's next 10,000 pages could cost less than its first 10.

Get a free API key

25 free credits · no card · read and map are always free

Cheap scraping at scale for AI agents.

Try it: no sign-up

Measured costs

How it gets cheap

Learn

Reuse

Repair

Cost per 1,000 pages vs. how often you revisit

When sites block the cloud

Rung 1: Our servers

Rung 2: Residential

Rung 3: Refuse with a reason

Receipts, not guesses

Use cases

Price & stock monitoring

RAG & docs ingestion

Research agents

Connect your agent in one line

MCP tools and what they cost

Pricing

How the prices compare

Questions

Your agent's next 10,000 pages could cost less than its first 10.