Best Web Scraping Tools 2026: 13 Compared, Free and Paid

String: the highest success rate and the fastest latency score

String is a web access API for developers and AI agents that returns the page at any URL from one HTTP call. Proxy rotation, anti-bot handling, captcha solving and JavaScript rendering happen behind the endpoint.

Pricing. Usage-based. Growth costs $99 a month (as of October 2026) and bills $0.16 per 1,000 fetches on a standard proxy, $0.90 on a premium proxy, $0.80 for a browser request on a standard proxy, and $3.25 with both. Starter costs $19 a month at $0.25, $2.40, $1.20 and $4.80. The first 20,000 standard fetches are free, blocked requests are not billed, and credits roll over. Protected sites can need a browser or a premium proxy, which bill at the higher rates.

Benchmark success. 98.6%, or 493 of 500, first of 17. All five attempts on 95 sites, partial content on 5, and nothing on none.

Latency. A latency score of 4.24 seconds, the fastest of 17.

Strengths. In the October 30, 2026 benchmark run, 100% against AWS WAF, Kasada, Fastly and in-house systems, 98.7% against Cloudflare, 97.9% against DataDome and 97.0% against Akamai. A hosted MCP server gives an agent fetch, search and sitemap discovery without local setup, and a jsonSchema field returns structured JSON with only the fields you define.

Weaknesses. In the October 30, 2026 benchmark run, 88.0% on grocery sites, with two of five on kroger.com, and 95.6% on news and finance, where Scrape.do, Scrapfly and Context.dev returned 100%.

Pair it with. A crawl loop of your own. String discovers a site's URLs through its sitemap endpoint and fetches each page; a framework such as Scrapy or Crawl4AI can drive the crawl and hand String the pages that block it.

Scrape.do: 94.0% and the lowest plain rate here

Pricing. As checked October 27, 2026, Pro is $99 a month for 1,250,000 credits, about $0.08 per 1,000. A datacenter request is 1 credit, the headless browser 5, a residential or mobile request 10, and residential with the browser 25: about $0.08, $0.40, $0.79 and $1.98 per 1,000. Domains listed in its request cost docs carry their own price, and those docs count 2XX, 400, 404 and 410 responses as successful, so a target's 404 bills.

Benchmark success. 94.0%, or 470 of 500, second of 17, on its first run. All five attempts on 91 sites, partial on 6, nothing on 3 (temu.com, shop.lululemon.com and asda.com).

Latency. A latency score of 5.08 seconds, second of 17.

Strengths. In the October 30, 2026 benchmark run, 100% against PerimeterX, AWS WAF, Kasada and Fastly, 97.9% against DataDome, and 100% on news and finance, social media, real-estate, review, jobs and marketplace sites. One GET endpoint, with ready-made scrapers for Amazon, Google and Walmart.

Weaknesses. In the October 30, 2026 benchmark run, 72.0% on grocery sites, and 87.0% against Akamai, its weakest wall. 16 of its 2xx responses were not the requested page. It publishes no MCP server of its own. See the Scrape.do alternatives.

Scrapfly: 86.8% and five APIs on one key

Pricing. As checked November 11, 2026, Pro is $100 a month for 1,000,000 credits, with overflow at $3.50 per 10,000. A datacenter scrape is 1 credit, JavaScript rendering adds 5, and residential proxies are 25, so a page costs from $0.10 to $2.50 per 1,000 and more with rendering. Failed requests are free, except 404s and except when more than 30% of an hour's traffic fails, per its billing docs. Credits do not roll over. The free tier is 1,000 credits with no time limit.

Benchmark success. 86.8%, or 434 of 500, fourth of 17. Nothing returned on 5 sites (Temu, Best Buy, Zara, Alibaba and Kroger), the third-fewest of the nine here after String and Scrape.do.

Latency. A latency score of 19.62 seconds, tenth of 17. On the 66 sites where Scrapfly and String both passed every attempt, Scrapfly's median was 4.94 seconds against String's 1.32.

Strengths. Scraping, browser, crawler, extraction and screenshot APIs share one key and one credit pool, and it ships a hosted MCP server. It holds SOC 2 Type II and ISO 27001 attestations and GDPR compliance on every plan; HIPAA (a BAA) and a DPA are Enterprise or Custom only. In the October 30, 2026 benchmark run, 92.6% against DataDome and 96.7% against AWS WAF, and 100% on news and finance, ticketing and developer sites.

Weaknesses. Speed on pages both return, and a bill that moves with the setting: its anti-bot mode can move a request to the residential pool, which is 25 times the datacenter rate. In the October 30, 2026 benchmark run, 40% on the one Fastly site.

ScraperAPI: 76.4%, and a fall from third to ninth

Pricing. As checked November 11, 2026, the plan nearest $100 is Startup at $149 a month for 1,000,000 API credits and 50 concurrent threads; Hobby is $49. A plain request is 1 credit. Premium proxies cost 10, ultra-premium 30, and requests to Cloudflare, DataDome, PerimeterX and Akamai sites cost 10 each, per its credit docs. It bills 200 and 404 responses. Credits do not roll over. The free plan is 1,000 credits a month, and a 7-day trial gives 5,000.

Benchmark success. 76.4%, or 382 of 500, ninth of 17, down from 84.0% and third place in September. All five attempts on 52 sites, partial on 39, nothing on 9, among them Facebook, Instagram and X.

Latency. A latency score of 15.19 seconds, sixth of 17.

Strengths. In the October 30, 2026 benchmark run, 81.0% against Akamai, second of the eight other APIs after Scrape.do, and 93.3% on developer sites. An MCP server.

Weaknesses. In the October 30, 2026 benchmark run, 60.0% on social media, and partial results on 39 of the 100 sites, second among the nine here to Bright Data's 40. Its August score was 64.2%; the September adapter switched on the setting ScraperAPI documents for protected pages, so that jump measured our adapter, not a product change. See String vs ScraperAPI and the ScraperAPI alternatives worth testing.

Context.dev: one flat credit per page

Pricing. As checked November 11, 2026, Developer is $20 a month for 5,000 credits and Pro is $90 for 100,000, or $0.90 per 1,000 pages, with overage at $2.00 per 1,000. One credit is one scrape, with proxies, rendering and anti-bot bypass included. Browser actions cost 2 credits. Failed requests are not billed, but a 404 bills at the base price. The free tier is 1,000 credits a month with no card.

Benchmark success. 93.0%, or 465 of 500, third of 17, up from 72.0% and eighth place in September. All five attempts on 91 sites, partial on 3, nothing on 6.

Latency. A latency score of 10.96 seconds, fourth of 17.

Strengths. The most predictable bill here, because a protected page costs the same credit as an easy one. In the October 30, 2026 benchmark run, 100% against DataDome, the only API in the run to return every DataDome request, and 100% on fashion, travel, news and finance, and social media sites. Against Firecrawl, its closest hosted rival on price per page, Context.dev returned more attempts on 18 sites and Firecrawl on 7; they tied on 75. An MCP server.

Weaknesses. In the October 30, 2026 benchmark run, 79.0% against Akamai, and nothing returned on 6 sites, among them Mouser, eBay and Kroger.

Firecrawl: crawl-to-Markdown and the third-fastest latency score

Pricing. As checked November 11, 2026, Standard is $99 a month billed monthly for 100,000 credits, or $0.99 per 1,000 pages; the page defaults to the annual view at $83. A scrape that returns no document is free, but a page that responds with a 403 or 404 is returned and costs a credit. Credits roll over on Scale and Enterprise; its billing docs limit that to annual plans. The free tier is 1,000 credits a month.

Benchmark success. 81.2%, or 406 of 500, sixth of 17. All five attempts on 80 sites, fourth of the nine here after String, Scrape.do and Context.dev. Nothing on 17, among them LinkedIn, Reddit, Instagram and TikTok.

Latency. A latency score of 8.87 seconds, third of 17.

Strengths. Crawl, map, search and scrape endpoints built for LLM input, an open-source core under AGPL-3.0, SOC 2 Type II, and an MCP server. In the October 30, 2026 benchmark run, 100% against PerimeterX and Kasada, 98.7% against Cloudflare, and 100% on real-estate and ticketing sites.

Weaknesses. In the October 30, 2026 benchmark run, 60.0% on in-house protection systems and 25% on social media. A Firecrawl engineer wrote in the adapter pull request that it blocks some sites by policy. See our Firecrawl alternatives page for the full breakdown.

Bright Data: proxies, datasets and unblocking under one contract

Pricing. As checked November 11, 2026, no plan near $100. Pay as you go is $1.50 per 1,000 successful requests, and the Scale tier is $499 a month for 383,000 requests. The free tier is 5,000 credits a month, shared across Web Unlocker API, SERP API, Web Scraper API, Scraper Studio and Browser API.

Benchmark success. 75.4%, or 377 of 500, tenth of 17. All five attempts on 49 sites, partial on 40, nothing on 11.

Latency. A latency score of 27.88 seconds, fifteenth of 17, against 15.62 seconds in September.

Strengths. The broadest catalogue here, with proxies, a browser API, SERP APIs and datasets, plus a free MCP server. In the October 30, 2026 benchmark run, 94.3% on real-estate sites and 93.3% on review and local sites.

Weaknesses. In the October 30, 2026 benchmark run, 45% on ticketing sites, 40% on search engines, and the slowest latency score of the nine here.

Oxylabs: bandwidth billing and 97.8% on news and finance

Pricing. As checked November 11, 2026, Web Unblocker is billed by bandwidth, on one Oxylabs subscription that also covers its proxies and scraping API. Starter is $15 a month at $7.20 per GB, Basic $100 at $6.00 per GB for up to 16GB, Flexible $500 at $4.80 per GB, and Enterprise $3.00 per GB. New accounts get a $5 welcome credit.

Benchmark success. 67.4%, or 337 of 500, thirteenth of 17. All five attempts on 61 sites, nothing on 26, among them Amazon, StubHub and Ticketmaster. The harness calls Web Unblocker without its JavaScript-render header; our Oxylabs page explains the drop from August.

Latency. A latency score of 16.62 seconds, seventh of 17.

Strengths. In the October 30, 2026 benchmark run, 97.8% on news and finance, 91.1% against PerimeterX, and 100% against Kasada. A hosted MCP server at mcp.oxylabs.io; the old GitHub repo was archived on September 16, 2026.

Weaknesses. Bandwidth billing makes heavy pages expensive and forecasts hard. In the October 30, 2026 benchmark run, 16.7% against AWS WAF, 47.3% on fashion, 0% on ticketing, and nothing on the one Fastly site.

Apify: a marketplace of prebuilt scrapers

Pricing. Free includes $5 (as of October 2026) of monthly usage, Starter is $19 a month and Scale is $199, each as prepaid usage plus pay as you go. Compute runs $0.20 per unit on Starter and $0.16 on Scale, and residential proxies cost $8 per GB on Starter. Unused prepaid usage expires at the end of each billing cycle.

Store Actors charge either per event or per usage, and each Actor's page says which. Web Fetch charges from $1.00 per 1,000 successful fetches plus a small fee per Actor start; a 404 it converts to Markdown counts as a fetch.

Benchmark success. 79.4%, or 397 of 500, eighth of 17, through the Web Fetch Actor. All five attempts on 65 sites, nothing on 12. Other Store Actors were not tested.

Latency. A latency score of 24.87 seconds, fourteenth of 17.

Strengths. The Apify Store lists 116,301 Actors, so a scraper for a named site often already exists. It offers scheduling, storage, an MCP integration, and Crawlee, its open-source crawling library. In the October 30, 2026 benchmark run, 100% against Kasada and on developer sites.

Weaknesses. Each Actor is maintained by its own developer, so quality and upkeep vary, and a site redesign can break one. Costs depend on run time and retries, which makes a bill hard to predict before a test run. See our Apify alternatives page.

Oxylabs vs Apify: which is better for scraping?

Oxylabs and Apify sell different things. Oxylabs Web Unblocker is a proxy endpoint that returns the page you ask for and bills by bandwidth. Apify is a platform of prebuilt scrapers, and its Web Fetch Actor does the same job as an unblocker: a URL in, the page out. On the October 2026 benchmark, Apify's Web Fetch Actor passed 79.4% of requests and Oxylabs passed 67.4%.

Site by site, Apify returned more attempts on 30 of the 100 sites and Oxylabs on 16. They tied on 54. Apify returned all five attempts on 65 sites and Oxylabs on 61. Oxylabs returned nothing on 26 sites and Apify on 12. Together they covered 79 sites in full.

Oxylabs led on Lowe's, Home Depot, Etsy, eBay, Coupang, Indeed, Reddit, TikTok, Pinterest, Reuters and WSJ. Apify led on Amazon, Zara, Nike, Kroger, Trustpilot, Yelp, StubHub, Ticketmaster, SeatGeek, StockX and Realtor.com. By protection in the October 30, 2026 benchmark run, Apify led on DataDome (85.3% to 67.4%), Cloudflare (86.7% to 68.0%), AWS WAF (63.3% to 16.7%) and Akamai (69.0% to 53.0%). Oxylabs led on PerimeterX (91.1% to 86.7%) and had the faster latency score, 16.62 seconds to 24.87.

Pick Oxylabs if you already buy its residential proxies and your list looks like the retail sites above. Pick Apify if you also want its Store scrapers and scheduling on the same account. Per their pricing pages, checked October 2026, Oxylabs Basic costs $100 for about 16GB a month; Web Fetch costs from $1.00 per 1,000 successful fetches on top of an Apify plan.

Crawl4AI: open-source crawling for language models

Pricing. Free. Crawl4AI is open source under Apache-2.0 and runs on your own machines. Any LLM extraction you add bills through your own model provider.

Benchmark success. Not tested. Its results depend on the proxies and browser settings you supply.

Strengths. Built to turn pages into clean Markdown for RAG and agents, with a self-hosting guide and a large, active community on GitHub. No per-page fee at any volume.

Weaknesses. No proxy pool and no unblocking service are included, so a DataDome or Akamai site blocks it the way it blocks any headless browser. You also run and patch the infrastructure.

Scrapy: the Python framework for large crawls

Pricing. Free. Scrapy is open source under BSD-3-Clause.

Benchmark success. Not tested.

Strengths. A long-established Python crawling framework, with asynchronous request scheduling, auto-throttling, item pipelines and middlewares, as its overview describes. It is fast on sites that serve plain HTML and it scales to very large crawls.

Weaknesses. It does not run JavaScript by default. Its own documentation on dynamically loaded content points to parsing embedded data or adding a headless browser. It has no proxies or captcha handling, so protected sites need an API or proxy service behind it. Several APIs on this page, including Scrapfly, publish Scrapy integrations for that reason.

Scrapy vs a hosted scraping API: when to switch

Stay on Scrapy while your sites serve plain HTML and retries fix the blocks you see. Add an API when the same 403, 429 or 503 comes back after retries, when a 200 carries a challenge page instead of the content, when the data loads by JavaScript, or when proxies and captchas take more of the week than the data. You do not have to replace Scrapy. Keep the spider for crawling and send only the blocked pages to the API.

This spider does that. It is one file, it runs with scrapy runspider, and the middleware retries a blocked response once through String's fetch endpoint:

"""Keep Scrapy for the crawl. Send only the pages a site blocks to the String Web Access API.

    pip install scrapy
    export STRING_API_KEY=...
    scrapy runspider hybrid_spider.py -o pages.jsonl
"""
import json
import os

import scrapy
from scrapy.http import HtmlResponse

STRING_FETCH = "https://request.usestring.ai/v1/fetch"
BLOCK_STATUSES = {401, 403, 429, 503}


class StringFallbackMiddleware:
    """Retry a blocked response once through String; pass every other response through."""

    def process_response(self, request, response, spider):
        if "string_for" in request.meta:
            page = json.loads(response.text)
            return HtmlResponse(
                url=request.meta["string_for"],
                status=page["statusCode"],
                body=page["data"].encode(),
                encoding="utf-8",
                request=request.replace(url=request.meta["string_for"]),
            )
        if response.status in BLOCK_STATUSES:
            spider.logger.info("blocked (%s), retrying through String: %s", response.status, request.url)
            return scrapy.Request(
                STRING_FETCH,
                method="POST",
                body=json.dumps({"url": request.url}),
                headers={
                    "Authorization": f"Bearer {os.environ['STRING_API_KEY']}",
                    "Content-Type": "application/json",
                },
                meta={**request.meta, "string_for": request.url},
                callback=request.callback,
                dont_filter=True,
            )
        return response


class PagesSpider(scrapy.Spider):
    name = "pages"
    start_urls = [
        "https://books.toscrape.com/",  # open site: plain Scrapy, no API call
        "https://www.stockx.com/",  # protected site: Scrapy gets a 403, String returns the page
    ]
    custom_settings = {
        "DOWNLOADER_MIDDLEWARES": {__name__ + ".StringFallbackMiddleware": 585},
        "HTTPERROR_ALLOW_ALL": True,
        "RETRY_ENABLED": False,
        "DOWNLOAD_TIMEOUT": 90,
        "ROBOTSTXT_OBEY": False,
    }

    def parse(self, response):
        yield {
            "url": response.url,
            "status": response.status,
            "via_string": "string_for" in response.request.meta,
            "title": response.css("title::text").get(default="").strip(),
            "bytes": len(response.body),
        }

We ran it on October 5, 2026 with Scrapy 2.13.4. books.toscrape.com came back through plain Scrapy with no API call. stockx.com returned HTTP 403 to Scrapy, and the middleware sent it through String, which returned the page and its title. Only the blocked page used an API request, and String does not bill a request it cannot return, so the fallback costs nothing on a page it also fails.

Scrapfly publishes a Scrapy integration for the same pattern. A 200 that carries a challenge page needs a content check as well as a status check; how to detect a bot block behind an HTTP 200 shows one.

Scrapy vs a hosted crawl API: which is better for a 100,000-page site?

Scrapy is better for a 100,000-page site whose pages are plain HTML and open to bots, and a hosted API is better when the site blocks plain clients. The software costs nothing either way, so the comparison is about who carries the blocking work.

String's rates are Growth rates as of October 2026, from String's pricing page. The time figure is the plan's request rate, 30 a second, applied to 100,000 requests. A site that rate-limits you will stretch a Scrapy crawl too, so check what the site allows before you pick a concurrency.

We did not benchmark Scrapy, because its result depends on the proxies and code you add. The benchmark covers the hosted APIs only. If you own a proxy pool and the site serves plain HTML, Scrapy's cost for 100,000 pages is mostly server time. If the site returns 403s after retries, the proxy and captcha work is the cost, and the API is cheaper in time. The middleware above gives you both: Scrapy crawls, and only the blocked pages go to the API.

Open-source scraping frameworks vs paid scraping APIs: which costs less?

An open-source framework costs less on sites that do not block you, and a paid API costs less on sites that do. The framework has no license fee, but you pay for servers, proxies, captcha solving and the hours spent fixing blocks. A paid API bills per page and carries that work.

On open sites, Scrapy or Crawl4AI fetches pages at the cost of your own compute, with no per-page fee at any volume. On protected sites, a framework returns the block page, and the fix is a proxy pool or an API. On the October 2026 benchmark of 100 bot-protected sites, the nine hosted APIs returned between 67.4% and 98.6% of requests, and none of the frameworks was tested.

At the plans nearest $100 a month, the APIs here run from about $0.08 to $0.99 per 1,000 standard pages, and from $0.40 to $3.25 per 1,000 with a browser or premium routing.

Run a framework when your sites are open, your team writes the spiders and you want no per-page bill. Buy an API when the same blocks keep coming back after retries. Many teams run both.

Which scraping APIs offer ready-made site integrations?

Scrape.do and Apify are the two on this page that sell ready-made scrapers for named sites. Scrape.do offers ready-made scrapers for Amazon, Google and Walmart behind its one GET endpoint, and its request cost docs list domains with their own price. Apify's Store lists 116,301 Actors, each built and maintained by its own developer.

That second fact answers the maintenance question for Apify: when a site changes, the Actor's developer fixes it, not Apify, so upkeep varies by Actor. Check an Actor's last update and its issues before you depend on it. For Scrape.do we did not verify who fixes a ready-made scraper when its site changes, so ask the vendor before you rely on one.

String, Context.dev, Scrapfly, ScraperAPI, Firecrawl, Bright Data and Oxylabs are generic fetch APIs on this page. Each returns the page you request, and your code or a schema extracts the fields, so a site redesign breaks your parsing and not a vendor-built scraper. Bright Data also sells datasets, which are collected data and not a scraper you run. String's jsonSchema field returns structured JSON with the fields you name, which keeps the extraction in your hands.

Playwright: browser automation for pages that need a real browser

Pricing. Free. Playwright is Microsoft's open-source browser automation framework under Apache-2.0.

Benchmark success. Not tested.

Strengths. One API drives Chromium, Firefox and WebKit, as its getting-started guide shows, and it can intercept network requests to read the JSON a page loads.

Weaknesses. It is a testing tool first. A browser alone brings no residential IPs and no fingerprint management, so protected sites detect and block it. At scale you run and pay for every browser.

Octoparse: no-code scraping

Pricing. As checked October 5, 2026, a free plan includes 10 tasks and 50,000 exported rows a month. Its pricing cards show Standard at $99 a month and Professional at $249; its FAQ quotes $69 and $249 billed annually, a 16% saving. Paid plans carry a 5-day money-back guarantee.

Benchmark success. Not tested.

Strengths. A desktop app where you click the data you want, with 500+ preset templates on Standard. Standard includes IP rotation, residential proxies, captcha solving and cloud runs.

Weaknesses. Standard allows 100 tasks; its card says up to 6 concurrent cloud processes and its FAQ says 3. Professional allows 250 tasks and 20. Click-built tasks break when a site changes its layout, and the app does not suit a pipeline that needs code review or version control.

The anti-bot cut

Success rates by protection vendor for the nine APIs, from the October 30 run. Site counts are in the column headers.

In the October 30, 2026 benchmark run, among the eight other APIs, Scrape.do leads or shares the lead on Akamai, PerimeterX, AWS WAF and in-house systems, Context.dev on DataDome, AWS WAF and in-house systems, and Firecrawl on Cloudflare and PerimeterX; five tie at 100% on Kasada. Two of those leads beat String: Context.dev's 100% on DataDome against 97.9%, and Scrape.do's and Firecrawl's 100% on PerimeterX against 97.8%. That spread is why no one of them covers a mixed list.

How many web scraping tools do you need? The stack count

Every page in this series carries one statistic computed from the run's per-site results. For a page about tools, the question is how many you have to buy.

Take the eight other APIs on this page: Scrape.do, Scrapfly, ScraperAPI, Firecrawl, Apify, Bright Data, Context.dev and Oxylabs. Count the sites where at least one of them returned all five attempts, for the best possible combination at each size.

• One tool: 91 of 100 sites in the October 30, 2026 benchmark run, from Scrape.do or Context.dev.
• Two tools: 98 of 100 in the benchmark run, from Scrape.do and Context.dev (Firecrawl in place of Scrape.do ties).
• Three tools: 99 of 100 in the benchmark run, adding Firecrawl (Firecrawl, Context.dev and Oxylabs tie).
• Four to eight tools: still 99. No further API adds a site.

In the October 30, 2026 benchmark run, String alone returned all five attempts on 95 of 100 sites. Two other APIs together pass it by three sites, and three by four. Each size adds less: 91, then 7 more, then 1.

The stack is real coverage. It also means several integrations, several sets of failure handling and several bills. At the plans nearest $100 a month (published prices, checked October 2026), the two-tool stack costs $189 in plans before usage (Scrape.do Pro $99 and Context.dev Pro $90, or Firecrawl Standard $99 in Scrape.do's place), and the three-tool stack $288 (adding Firecrawl Standard $99). You also need routing code that knows which API to send each site to, and that code breaks when a site changes protection.

One site returned in full through String and through no combination of the others: temu.com. Five sites return in full through the stack and not through String: kroger.com, where String returned two of five, and skyscanner.net, yellowpages.com, marketwatch.com and barrons.com, where String returned four of five.

Why pricing is hard to compare across web scraping tools

1. The unit is different. String, Scrape.do, Firecrawl and Context.dev price per request or credit. Oxylabs prices per gigabyte. Apify prices compute units and events. Octoparse prices tasks. The frameworks are free, and the proxies you buy for them are the real cost.
2. A credit is not a page. One Context.dev credit is one scrape. One Scrapfly credit is a datacenter fetch, and the same page on residential proxies is 25. One ScraperAPI request to a Cloudflare, DataDome, PerimeterX or Akamai site is 10 credits.
3. Failure billing differs. String and Bright Data bill only successes. Scrapfly bills failures when more than 30% of an hour's traffic fails, and 404s (vendor billing docs, checked October 2026). Context.dev and Scrape.do bill 404s. Firecrawl bills a 403 or 404 page that returns a document, and ScraperAPI bills 404s.
4. Annual prices pose as monthly. Firecrawl shows $83 by default and bills $99 monthly. Octoparse's FAQ quotes $69, the annual rate (vendor pricing pages, checked October 2026).
5. Unused credit often expires. String's credits roll over. Scrapfly's do not, Apify's prepaid usage expires monthly, and Firecrawl rolls over only on annual Scale and Enterprise. ScraperAPI's and Context.dev's do not roll over.

What $100 a month buys

• String, $99 (as of October 2026): $0.16 per 1,000 standard fetches, $0.90 premium, $0.80 browser, $3.25 browser on premium. Failed requests free, credits roll over.
• Scrape.do, $99: 1,250,000 credits, or about $0.08 to $1.98 per 1,000 requests by mode. A target's 404 bills (pricing page, checked October 27, 2026).
• Scrapfly, $100: 1,000,000 credits, or $0.10 to $2.50 per 1,000 pages depending on setting. No rollover (pricing page, checked November 11, 2026).
• ScraperAPI: no plan at $100; Startup at $149 for 1,000,000 credits, 10 per protected-site request (pricing page, checked November 11, 2026).
• Firecrawl, $99: 100,000 credits, or $0.99 per 1,000 pages (pricing page, checked November 11, 2026).
• Context.dev, $90: 100,000 credits, or $0.90 per 1,000 pages (pricing page, checked November 11, 2026).
• Bright Data: no plan near $100; $1.50 per 1,000 successful requests pay as you go (pricing page, checked November 11, 2026).
• Oxylabs, $100: about 16GB of Web Unblocker bandwidth at $6.00 per GB (pricing page, checked November 11, 2026).
• Apify: no $100 plan; $19 or $199 of prepaid usage, Web Fetch from $1.00 per 1,000 (pricing page, checked November 11, 2026).
• Octoparse: Standard at $99 on its pricing page, checked October 5, 2026.
• Crawl4AI, Scrapy, Playwright: free software; your servers and proxies are the bill.

Web scraping MCP servers and what they cost

Every hosted API on this page except Scrape.do publishes an MCP server, so an agent can call it as a tool. None of them charges extra for the MCP route. A call bills the same account and the same rate as the API.

So the cheapest MCP server is the cheapest API for your sites. As of October 2026, at $99 a month, String bills $0.16 per 1,000 standard fetches and nothing for a failed one, which matters for an agent that retries.

Which web scraping tool should you pick?

• Your sites are protected and the pages have to come back. String, at 98.6% on the benchmark and the fastest latency score.
• You want the lowest plain rate on a high-scoring API. Scrape.do, at 94.0% on the October 2026 benchmark and about $0.08 per 1,000 datacenter requests (pricing page, checked October 27, 2026).
• You want one predictable number in the budget. Context.dev, one credit per page.
• You crawl sites into Markdown for an LLM. Firecrawl for a hosted service, Crawl4AI if you want to run it yourself.
• You want many APIs under one key. Scrapfly.
• Your hard sites run DataDome. Context.dev, at 100% against DataDome on the October 2026 benchmark.
• Your list is news and finance. Scrape.do, Scrapfly or Context.dev, all at 100% on the October 2026 benchmark, then Oxylabs at 97.8%.
• You need proxies, datasets and SERP data on one contract. Bright Data.
• Someone already built a scraper for your site. Apify, after a test run to price it.
• Your sites are open and your team writes Python. Scrapy. Writes TypeScript: Crawlee.
• You already run Scrapy and some sites block it. Keep Scrapy and add the fallback middleware above.
• Your pages need clicks or logins. Playwright, with an unblocking API behind it for protected sites.
• You do not write code. Octoparse.
• Your list is retail, marketplaces or grocery. Pick from our best e-commerce scraping APIs, which scores the same APIs on the 43 e-commerce sites alone. Overall rank is a poor guide there: the median provider's score swings 22.8 points between its best and worst kind of shop.

Which web crawl MCP tools are priced well?

The best-priced web crawl MCP tools are the cheapest scraping APIs for your sites, because each hosted API on this page bills MCP calls at its normal API rate. String's hosted MCP server bills $0.16 per 1,000 standard fetches on the Growth plan, as of October 2026, per String's pricing page.

Which web scraping tool has the best performance test?

String has the best performance test result: it returned content on 98.6% of requests, first among the hosted APIs tested, in String's Web Data Frontier Benchmark (October 30, 2026 run). String built the benchmark, and its code is public so you can rerun it with your own keys.

What web crawling platforms are worth comparing for commercial use?

For commercial use, compare the hosted APIs that handle blocking and bill per page: String, Scrape.do, Context.dev, Scrapfly, Firecrawl, Apify, ScraperAPI, Bright Data and Oxylabs. The best of them returned content on 98.6% of requests in String's Web Data Frontier Benchmark (October 30, 2026 run), so test a shortlist on your own sites.

FAQ

What are the best web scraping tools for AI agents?

For an agent that must fetch live pages, a hosted API with an MCP server is the shortest path, because the agent calls one tool and gets the page back. String passed 98.6% of requests on the October 30, 2026 benchmark, with the fastest latency score, and ships a hosted MCP server that bills at the API rate. Scrapfly, ScraperAPI, Firecrawl, Apify, Bright Data, Context.dev and Oxylabs also publish MCP servers. Crawl4AI is the open-source option when you run your own infrastructure and your sites are not heavily protected.

How many web scraping tools do I need?

For a list of bot-protected sites, more than one API can help, but each one adds less. On our 100-site benchmark the best single API other than String returned all five attempts on 91 sites, the best pair on 98 and the best three on 99. String alone returned 95. The two-tool stack costs $189 a month in plans at the $100 anchor, before usage and routing code.

What is the best no-code web scraping tool?

Octoparse is the established no-code option. Its free plan includes 10 tasks and 50,000 exported rows a month, and Standard costs $99 a month on its pricing page (checked October 5, 2026), with residential proxies and captcha solving included. We did not benchmark it. Its limits are task counts and concurrency, and click-built tasks need fixing when a site changes its layout.

Are open-source web scraping frameworks better than paid scraping APIs?

They answer different problems. Scrapy, Crawlee, Playwright and Crawl4AI are free and give full control, and they work well on sites that do not block automation. A paid API earns its price on protected sites, because it supplies the residential proxies, browser fingerprints and captcha handling a framework leaves to you. On our benchmark's 100 protected sites, the nine APIs on this page returned between 67.4% and 98.6% of requests.

Scrapy vs a hosted scraping API: when should I switch?

When retries stop fixing blocks, when pages need a browser, or when proxy and captcha work takes more time than the data. You can keep Scrapy for the crawl and send only blocked pages to an API; the section above has a working middleware.

Managed scraping service vs scraping API: which do I need?

A scraping API returns the pages you request, and your code decides what to extract. A managed service builds and runs the collection and delivers the data. Our managed scraping service vs scraping API page covers the choice, the SLA to ask for and the cost.

Which web scraping tools have a free tier?

String gives the first 20,000 standard fetches free. Scrapfly gives 1,000 credits and ScraperAPI 1,000 a month, and ScraperAPI adds a 7-day, 5,000-credit trial. Firecrawl gives 1,000 credits a month, Context.dev 1,000 a month, and Bright Data 5,000 credits a month. Apify's free plan includes $5 of monthly usage and Octoparse has a free plan with 10 tasks (vendor pricing pages, checked October and November 2026). Scrapy, Crawlee, Playwright and Crawl4AI are free open-source software.

Can I verify these benchmark numbers myself?

Yes. The harness, the 100 sites, the pass rule and every adapter are open on GitHub, and it runs with your own API keys. Per-site results for all 17 providers are on the benchmark page, and our three-way comparison of ScraperAPI, ZenRows and ScrapingBee shows how to test a shortlist on your own list.

Sources

Checked October 5, 2026; Oxylabs, Context.dev, ScraperAPI and Bright Data November 11, 2026; Scrape.do October 27, 2026. Benchmark figures: October 30, 2026 run, official_results/benchmark-2026-10-30T18-48-18-489Z.json in the harness repository.

Benchmark data

1. https://usestring.ai/benchmark
2. https://usestring.ai/blog/web-data-frontier-benchmark-october-2026
3. https://usestring.ai/blog/web-scraping-benchmark-problem
4. https://github.com/usestring/web-data-frontier-benchmark
5. https://github.com/usestring/web-data-frontier-benchmark/pull/6
6. https://github.com/usestring/web-data-frontier-benchmark/blob/main/official_results/benchmark-2026-10-30T18-48-18-489Z.json
7. https://github.com/usestring/web-data-frontier-benchmark/pull/23

Hosted APIs

8. https://usestring.ai/pricing
9. https://portal.usestring.ai/docs/get-started/pricing
10. https://portal.usestring.ai/docs/mcp/overview
11. https://portal.usestring.ai/docs/fetch/structured-extraction
12. https://scrape.do/pricing/
13. https://scrape.do/documentation/request-costs/
14. https://scrapfly.io/pricing
15. https://scrapfly.io/docs/scrape-api/billing
16. https://scrapfly.io/compliance
17. https://scrapfly.io/products/mcp-cloud
18. https://scrapfly.io/docs/scrape-api/anti-scraping-protection
19. https://scrapfly.io/docs/sdk/scrapy
20. https://www.scraperapi.com/pricing/
21. https://docs.scraperapi.com/getting-started/quick-start/credits-and-requests-costs
22. https://docs.scraperapi.com/integrations/llm-integrations/mcp-server
23. https://www.context.dev/pricing
24. https://docs.context.dev/install-mcp
25. https://www.firecrawl.dev/pricing
26. https://docs.firecrawl.dev/billing
27. https://docs.firecrawl.dev/mcp-server
28. https://github.com/firecrawl/firecrawl
29. https://brightdata.com/pricing/web-unlocker
30. https://brightdata.com/ai/mcp-server
31. https://oxylabs.io/products/web-unblocker/pricing

Marketplace, frameworks and no-code

32. https://apify.com/pricing
33. https://apify.com/apify/web-fetch
34. https://apify.com/store
35. https://docs.apify.com/platform/actors/running/actors-in-store
36. https://docs.apify.com/platform/integrations/mcp
37. https://github.com/apify/crawlee
38. https://github.com/apify/crawlee-python
39. https://github.com/unclecode/crawl4ai
40. https://docs.crawl4ai.com/
41. https://github.com/scrapy/scrapy
42. https://docs.scrapy.org/en/latest/intro/overview.html
43. https://docs.scrapy.org/en/latest/topics/dynamic-content.html
44. https://books.toscrape.com/
45. https://github.com/microsoft/playwright
46. https://playwright.dev/docs/intro
47. https://playwright.dev/docs/network
48. https://github.com/puppeteer/puppeteer
49. https://github.com/SeleniumHQ/selenium
50. https://www.crummy.com/software/BeautifulSoup/
51. https://www.octoparse.com/pricing

Related String pages

52. https://usestring.ai/answers/detect-bot-block-http-200
53. https://usestring.ai/answers/managed-scraping-service-vs-scraping-api
54. https://usestring.ai/comparisons/string-vs-scraperapi