Free Web Scraping APIs (and Their Real Limits)

“Free” does four different jobs on a scraping API pricing page, and only one of them runs out. Recurring credits, self-hosted software, and keyless endpoints stay free; the one-time signup balance runs dry mid-project, which is when you notice the pricing page was pointing at the paid button all along. I checked every free tier against its own pricing page today (October 16, 2026), ran the keyless ones from this machine, and wrote down where each one stops, because a search for a web scraping API free tier should not end at a trial button.

One disclosure first. Magpie has no hosted API, so this page is not an ad for one. It’s a local CLI, and later I’ll say plainly where it fits and where it doesn’t. The hosted entries stand on their own pricing.

Scraper API free tiers at a glance

Ranked by how long the free part lasts: monthly-recurring tiers first, one-time balances after, free-forever self-hosted last. Prices come from each vendor’s own pricing page, checked October 16, 2026; the paid-only rankings live on the main APIs list.

Tool	The free part	Refreshes?	Best for	The real limit
ScraperAPI	1,000 credits/month (5,000 in your first 7 days)	Yes, monthly	Small recurring jobs with proxy fallback	5 concurrent requests; hard sites burn credits fast
Firecrawl	1,000 credits/month, no card	Yes, monthly	URL/crawl → LLM-ready Markdown	One 1,000-page crawl spends the whole month; JSON formats cost +4/page
SerpApi	250 searches/month	Yes, monthly	Google/Bing SERP data only	Not a page scraper; search results only
Apify	$5 of platform credits/month	Yes, monthly	Marketplace actors (prebuilt scrapers)	Actor charges stack on top of platform credits; balances expire monthly
Jina Reader	Keyless, ~20 requests/min	Continuous	One-shot URL → Markdown from scripts	Markdown only; no crawl, no structured extraction
ScrapingBee	1,000 credits, no card	No, one-time	Testing whether a managed API fits	After that, $49/mo; stealth-proxy requests cost 75 credits each
Crawlbase	1,000 credits, no card	No, one-time	A day of testing two API styles	Signup balance, not an allowance
ScrapingDog	100 credits, no card	No, one-time	A single afternoon’s evaluation	Smallest balance on this page
Crawl4AI	Everything, forever (Apache-2.0)	n/a, self-hosted	Unlimited volume, full control	You own the server, the proxies, and the maintenance

1. ScraperAPI: the free tier that behaves like a plan

ScraperAPI’s free tier is the honest kind: 1,000 API credits that refresh every month, no expiry trick, plus a 5,000-credit first week if you want to test at a slightly real scale. Every request goes through their proxy pool, and JS rendering or premium proxies multiply the credit cost per request. Their docs say so upfront, which is more than most manage.

curl "https://api.scraperapi.com/?api_key=$SCRAPERAPI_KEY&url=https://example.com"

Pros:

• 1,000 credits/month recurring; a small daily job fits forever
• 5,000 bonus credits in the first 7 days
• Proxy rotation and retries handled server-side
• No credit card needed to start

Cons:

• 5 concurrent requests max on free, so parallel crawls queue
• Premium proxy or JS-render requests can eat 10–25+ credits each
• Past 1,000/month you’re into paid plans or pay-as-you-go

2. Firecrawl: 1,000 credits a month, tuned for LLM pipelines

Firecrawl turns a URL or a whole crawl into clean Markdown that feeds directly into RAG or agents. The free plan is 1,000 credits, refreshed monthly, no card, and one credit is one page on a basic scrape. That math is the whole review: a 1,000-page site crawl spends the month in one job. The JSON, question-answer, and highlight output formats add 4 credits per page, so a structured 100-page pull costs 500.

curl -X POST https://api.firecrawl.dev/v1/scrape \
  -H "Authorization: Bearer $FIRECRAWL_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com","formats":["markdown"]}'

The same engine is open source, so the free tier has a floor: self-host it and the credit meter disappears (with feature cuts, covered in the full Firecrawl breakdown).

Pros:

• 1,000 credits/month, no card, real monthly refresh
• Crawl + Markdown + structured extraction in one API
• Open-source core you can self-host

Cons:

• One full-site crawl exhausts the free month
• JSON/question formats bill 4 extra credits per page
• No credit rollover on entry paid plans (Hobby $19/mo for 5,000)

3. SerpApi: free searches, but only search results

SerpApi’s free plan gives 250 searches a month. It was 100 for years, and the raise is real, straight from their pricing page today. Only successful searches count; cached, errored, and failed ones don’t. But know what the product is: structured search-engine results pages, not web pages. If your “scraping” is actually “get me the top 20 Google results with sitelinks and answer boxes,” this is the cleanest free tier on the board. If you need page content, it does nothing for you.

curl "https://serpapi.com/search.json?engine=google&q=laptop+stands&api_key=$SERPAPI_KEY"

Pros:

• 250 searches/month free, no card
• Failed and cached searches don’t bill
• Every SERP feature parsed into JSON

Cons:

• SERPs only; no arbitrary page scraping
• Paid starts at $25/month for 1,000 searches
• Volume pricing climbs fast past that

4. Apify: $5 a month forever, spent at the marketplace

Apify’s free plan deposits $5 of platform credits every month, indefinitely, no card. You spend it in the actor marketplace: a Web Scraper actor, a Google Maps actor, an Instagram actor someone else maintains. The catch is that platform credits and actor pricing stack, so an actor that bills per result drains your $5 faster than the compute alone. Our Apify alternatives page ran the arithmetic: $5 buys roughly 1,250 Google Maps places, about $4 per thousand.

Pros:

• $5/month recurring, forever, no card
• Thousands of prebuilt actors, and you write none of them
• Scheduling and storage included

Cons:

• Actor charges stack on platform charges; credits expire monthly
• $5 is a demo budget, not a production one
• The heaviest actors finish the $5 in a day

5. Jina Reader: free with no key at all

Jina’s Reader is the only entry here you can use this second, with no signup: prefix any URL with r.jina.ai/ and get Markdown back. Jina documents the keyless limit around 20 requests per minute; a free key raises it to 100 RPM. I tested it today from a home connection: 25 distinct pages in 69 seconds (about 22/minute), every request a 200, roughly 2.8 seconds per page. That’s a real tool for scripts and agents that need one page at a time.

$ curl https://r.jina.ai/https://en.wikipedia.org/wiki/Web_scraping
Title: Web scraping

URL Source: https://en.wikipedia.org/wiki/Web_scraping

Markdown Content:
**Web scraping**, **web harvesting**, or **web data extraction** is data scraping used for extracting data from websites.

(Verbatim from today’s run, abridged.) The limits are what they look like: single-URL reads, Markdown only, no crawl, no schedule. And on the keyless tier you’re last in line when the shared infrastructure gets crowded.

Pros:

• No key, no card, no signup; curl and go
• Markdown output tuned for LLMs
• 100 RPM free with a (free) key

Cons:

• Single-URL reads only, no crawling, no scheduling
• No structured extraction, no JSON schema output
• Keyless tier shares infrastructure; no SLA, so don’t build a business on it

6. ScrapingBee: 1,000 credits, then the meter starts

ScrapingBee’s trial is 1,000 free API credits, no card. Enough to learn the API and scrape a few hundred easy pages. What the trial doesn’t show you is the credit multiplier table: simple requests cost 1 credit, but pages that need their stealth proxy cost 75. Their own help center publishes those numbers, and our ScrapingBee alternatives page did the division: the $49 Freelance plan’s 250,000 credits shrink to roughly 3,300 stealth pages a month.

Pros:

• 1,000 free credits, no card
• Simple REST shape, JS rendering with a parameter
• Decent docs and a real proxy pool

Cons:

• One-time balance; it does not refresh
• 75-credit stealth requests make “free” land differently
• Entry plan $49/month after

7. Crawlbase: 1,000 credits for a test day

Crawlbase offers 1,000 free credits with no card, spanning both of its products: a scraper API and a “crawler” that pushes results to a webhook you register. The free balance is a signup grant, not a monthly allowance, so treat it as evaluation hardware. One afternoon decides whether their two-API split fits your architecture.

Pros:

• 1,000 free credits, no card
• Push-based crawler mode (results via webhook) is genuinely useful
• Both API styles draw on the same balance

Cons:

• One-time balance, no monthly refresh
• Different request types drain the balance at different rates
• Paid tiers are built around much larger volumes

8. ScrapingDog: 100 credits for one afternoon

ScrapingDog currently hands out 100 free credits, no card. That’s the smallest balance on this page, and their own copy pitches it as a way to “spin the API,” which is exactly right. One sitting: render a few JS-heavy pages, check response quality, decide. If your evaluation needs more than 100 requests, the trial ends before your questions do.

Pros:

• No card required
• Simple GET-shaped API with a headless-browser parameter
• Fast to evaluate; the whole trial fits in an hour

Cons:

• 100 credits is a test drive, nothing more
• One-time, no refresh
• Paid plans start higher than some peers

9. Crawl4AI: the free tier with no meter at all

Crawl4AI is open source (Apache-2.0, 85,000+ GitHub stars) and does what the hosted APIs do, which is fetch, render JavaScript, clean to LLM-ready Markdown, and extract structured data. The difference is that it runs on your machine and costs nothing per request. It ships as a Python library and a Docker server with a REST API, so you can point scripts at localhost and pretend you have a hosted API with infinite credits.

pip install crawl4ai && crawl4ai-setup
# or run it as a local REST API:
docker run -d -p 11235:11235 --shm-size=1g unclecode/crawl4ai

The price moves to you. You own the rendering browser, the rate limiting, the updates, and the fact that your home IP is the proxy pool. For scraping politely at moderate volume, that’s a fine trade. For hitting protected targets at scale, it isn’t: no open-source library includes a rotating residential network. More free-forever options sit on the open-source scraping tools list.

Pros:

• Free forever, Apache-2.0, no credits, no key
• REST server mode mimics a hosted API locally
• Deep crawling, JS rendering, and CSS/XPath extraction built in

Cons:

• You run it: server, browser, memory, updates
• No proxy network, so protected sites block your IP instead of theirs
• Setup is minutes, not seconds, and debugging is yours

Web scraping API free plans: what you actually get

A free web scraping API tier is real in one of two shapes: a recurring monthly allowance or a one-time grant. The recurring ones (ScraperAPI, Firecrawl, SerpApi, Apify) survive a second month; the one-time ones (ScrapingBee, Crawlbase, ScrapingDog) are trials wearing the word “free.” Neither shape survives contact with protected targets at scale, for three reasons.

First, credit multipliers. Managed APIs charge 1 credit for an easy page and 10–75 for a page behind a bot wall, because the credits are proxy hardware wearing a unit label. The free tier covers the easy pages; the hard ones are what you actually needed. The full credit arithmetic is its own subject, and it’s the difference between a free month and a $49 surprise.

Second, ceilings. Five concurrent requests (ScraperAPI) or ~20 per minute keyless (Jina) is fine for a daily sync and hopeless for a one-off 5,000-page migration.

Third, no SLA. Free tiers sit on shared infrastructure with no uptime promise. Fine for a dashboard you check at breakfast; wrong for anything with a stakeholder.

One more line, because it belongs here: none of these tools should be used to defeat logins or terms of service you’ve agreed to. Hosted APIs vary in how hard they lean on that line. The polite approach is documented in scraping without getting blocked, and it’s mostly slowing down.

The best free scraping API might be the site’s own

Before spending any credits, spend one curl. Many sites already ship a free JSON API, and it beats every scraper on freshness, stability, and terms. Two from this machine, no keys, no signup:

$ curl "https://hn.algolia.com/api/v1/search?query=web+scraping&tags=story&hitsPerPage=2"
{"hits":[{"title":"Congrats! Web scraping is legal! (US precedent)", ...}]}

$ curl "https://en.wikipedia.org/api/rest_v1/page/summary/Web_scraping"
{"type":"standard","title":"Web scraping","wikibase_item":"Q665452", ...}

(Both abridged.) Hacker News, Wikipedia, GitHub, Reddit, YouTube, government data portals: most expose structured endpoints that never bother a polite client. Check example.com/sitemap.xml, open the browser network tab and look for fetch calls ending in .json, and search the footer for an API link before you rent someone’s proxy pool. This is also the honest first move for agent-driven scraping: with an API endpoint, the LLM step becomes optional.

Where Magpie fits (and where it doesn’t)

Magpie is not a hosted API. No account, no endpoint, no credits. It’s a local-first scraper: a single Go binary whose CLI you call like an API from scripts, free under AGPL-3.0-or-later (or a commercial license). If what you wanted from a hosted scraping API was the data, meaning fetch, clean, Markdown or JSON, it does that on your machine with zero signups:

$ magpie scrape --format json https://news.ycombinator.com
{
  "url": "https://news.ycombinator.com",
  "title": " Hacker News",
  "markdown": "|  1. | [Show HN: ...](item?id=...) ..."
}

3.0 seconds, 19KB of JSON, 18,836 characters of clean Markdown (abridged above; the leading space in the title is HN’s own, not a typo). For AI agents, magpie serve runs a local MCP server with 13 tools, including scrape_url, crawl_site, and extract_structured, so Claude or Cursor calls your machine the way they’d call a hosted API, except the requests never leave your network. magpie init --client claude-code writes the config stanza.

The limits, stated plainly: local means one machine, your IP, no proxy network, and protected targets will block you the same way they’d block Crawl4AI. It detects bot challenges and reports them; it does not solve them. If you want a hosted endpoint with a dashboard, use Firecrawl or ScraperAPI and pay when the free tier runs out. The cloud vs local comparison covers that trade in full, and the Desktop app’s pricing (5 free runs, then $49/month flat) is on the free scrapers list if you’d rather have a GUI than a CLI.

Free-tier FAQ

Is there a truly free web scraping API?

Truly free means no meter and no month. Jina Reader’s keyless endpoint is free per-request but rate-limited; Crawl4AI is free forever but you host it. Among hosted plans, ScraperAPI’s and Firecrawl’s 1,000 monthly credits are the most generous recurring offers: free within their caps, paid beyond them.

Which web scraper API is free forever?

Crawl4AI, self-hosted. Apache-2.0 licensed, no credit system at all. Firecrawl’s open-source server is the second answer: same engine, no cloud bill, with the feature cuts self-hosting implies. Every metered hosted API on this page charges eventually, by design.

Is a free scraper API good enough for production?

For under ~1,000 simple pages a month, yes. ScraperAPI’s or Firecrawl’s recurring tiers run small production jobs indefinitely. It stops being enough when targets sit behind bot protection (multiplied credit costs), when you need concurrency beyond the free ceiling, or when downtime becomes a business problem.

Do these free plans require a credit card?

No. All nine entries here start without one; that’s why they made the list. Jina’s keyless tier doesn’t even need an account. Cards enter the picture at pay-as-you-go, and ScraperAPI notably makes overage billing opt-in. Read that toggle before it bills you.

What happens when the free credits run out?

The API returns a quota error, usually 401 or 429, until the month resets or you pay. No data loss, no deletion; your account and history stay. The failure mode that surprises people isn’t the hard stop, it’s the slow one: multipliers draining the balance two weeks early on a “protected” target.

Pick by shape, not by brand

One-off bulk extraction where volume is everything: Crawl4AI on your own hardware, zero meter. A small job that must run every month without attention: ScraperAPI’s or Firecrawl’s recurring 1,000. Search results specifically: SerpApi’s 250. Testing before committing: Jina Reader, keyless, this afternoon. And check whether the site already has an API before any of that. The fastest free scraping API is the one with a docs page and no credits anywhere on it.