ScrapingAnt vs scrape.do: Pricing, Credit Costs and API Features

Quick verdict and provider profiles

For a small application that needs ordinary HTML and occasional browser rendering, begin with ScrapingAnt if its $19 commitment fits your budget. For recurring rendered or residential workloads, compare scrape.do’s lower mode multipliers before deciding. For supported Amazon, Google or other structured targets, a Ready API can reduce parser maintenance enough to outweigh the difference in subscription price.

Both providers handle retrieval infrastructure; your team still owns validation, deduplication, durable storage and most application scheduling. Our web scraping tools guide explains how these APIs compare with visual tools and hosted workflow platforms.

LOWER ENTRY COST · AI AND MARKDOWN

ScrapingAnt

ScrapingAnt suits developers who want browser-backed web access inside an existing application. Its AI extractor can describe fields rather than maintain selectors, with extra consumption based on text size. Read our ScrapingAnt review for its API modes and integration details.

Managed Chromium with custom JavaScript and selector waits

HTML, Markdown, extended responses and optional AI JSON

MCP, Make, n8n and GitHub Action integration routes

Explore ScrapingAnt plans

LOWER MODE MULTIPLIERS · ASYNC JOBS

scrape.do

scrape.do suits teams with recurring retrieval volume or a target covered by its structured catalog. Its generic API also supports HTML and Markdown; browser JSON has a different purpose from parsed Ready API fields. Our scrape.do review covers those distinctions.

Datacenter, residential/mobile routing and browser interactions

Ready APIs for supported search, ecommerce and other targets

Separate async concurrency, webhooks and structured plugins

Explore scrape.do plans

Plans, free access and billing conditions

All prices below are published US dollar API subscription prices before tax. Free offers are recurring credit allowances, not unlimited scraping. One response can consume several credits, and a parsed record may require multiple pages. Keep API subscriptions separate from standalone proxy products billed by bandwidth.

ScrapingAnt Web Scraping API plans

The technical error reference limits Free to one concurrent request and returns 409 when that is exceeded. The homepage uses broader no-cap wording for all plans; use the documented Free limit when sizing a pilot. Paid plans should still use a bounded queue and measured target limits. Startup adds custom integration snippets; Business tiers add debugging and account assistance, which can matter when a particular site needs tuning.

scrape.do API plans

Core rendering, browser interactions, geo-targeting, sessions and datacenter/residential/mobile routing are advertised across plans, including Free. Business separately lists premium residential/mobile options, teams and dedicated support; ordinary residential access is not Business-exclusive. Advanced advertises custom WAF assistance, an SLA and Slack support. Enterprise capacity and service commitments need a written contract.

Annual billing: compare the actual invoice

The published scrape.do annual totals are $290 for Hobby, $990 for Pro, $2,490 for Business and $6,990 for Advanced. Each equals ten monthly payments for twelve months: calculated savings are 16.67% against twelve month-to-month invoices. Monthly equivalents are $24.17, $82.50, $207.50 and $582.50. Its headline “up to 20%” should not replace that arithmetic. Annual arrangements are handled through support; confirm allocation timing, renewal and cancellation before paying.

No numeric ScrapingAnt annual discount is assumed here. A custom proposal can differ from public monthly rates. Start with monthly billing while establishing usage, particularly if your targets require expensive residential rendering or AI extraction.

Credits, renewal and refunds

ScrapingAnt’s unused API credits do not roll over. Its paid plan can be restarted mid-cycle by making another payment, refreshing the allowance. Monthly subscriptions renew automatically; cancellation needs to precede the next period. Published terms treat payments as non-refundable, with exceptions at the provider’s discretion.

scrape.do requests stop at the monthly allowance unless you upgrade or buy additional credits. Confirm rollover, purchased-credit expiry and annual credit allocation directly rather than assuming they are unlimited or permanent. Subscriptions auto-renew and can be canceled through the account or support. Its initial refund window is three days, covers unused credits and can deduct payment commissions; successfully used credits are excluded. Separate forgotten-renewal provisions describe full refunds within three days and 50% within fifteen days, subject to the stated conditions. A refund is not the same operation as canceling renewal.

Credit costs and realistic workload budgets

A lower subscription bill does not automatically mean a lower cost per useful response. ScrapingAnt’s browser rendering is enabled by default, while scrape.do uses an explicit render option for ordinary browser work. Specify the same proxy and rendering requirements before comparing charges.

ScrapingAnt also has browser-source modes: 2 credits with datacenter routing and 50 with residential routing when the browser is enabled but page source is returned without full JavaScript execution. These are different from its ordinary static and fully rendered modes. A Google-domain datacenter request costs 10 credits. Log the Ant-credits-cost header to reconcile what your calls actually consume.

scrape.do applies special pricing to some domains and can automatically select routing or rendering. Published examples include Amazon at 1 credit through its Amazon API route, Google at 10, LinkedIn at 30, Air Canada at 75 and Carsales at 25–200 depending on complexity. A more expensive requested profile can raise the charge. The Scrape.do-Request-Cost and Scrape.do-Remaining-Credits headers are more reliable for accounting than assuming every successful call costs one credit.

Nominal cost per thousand ordinary responses

These allocated figures assume the entire monthly credit pool is used. If you only fetch 1,000 simple pages in a month, the actual entry subscription outlay remains $19 or $29, not $0.19 or $0.116. Conversely, a large residential browser workload can exhaust the lower-cost subscription quickly. Pick a plan around expected total consumption and peak concurrency, rather than dividing the headline price by credits you will leave unused.

Example: a mixed retrieval batch

An illustrative batch of 5,000 static datacenter calls, 500 rendered datacenter calls and 40 rendered residential calls contains 5,540 responses. It consumes 15,000 ScrapingAnt credits or 8,500 scrape.do credits using ordinary-domain rules. At full entry-plan utilization, allocated costs are $2.85 and $0.986 respectively. If this is your only monthly work, the cash invoices remain $19 and $29. Named domains, AI processing, retries that become billable and extra pagination can change the result.

ScrapingAnt AI extraction adds a character-based charge

The AI extractor adds ceiling((input Markdown characters + output characters) ÷ 30) credits to retrieval. For 9,000 input characters, 600 output characters and a 10-credit browser fetch, the total is 330 credits. An Enthusiast allowance fits 303 complete requests at that assumed size, and the allocated full-use cost is $0.0627 each. The plain rendered-HTML capacity of 10,000 responses is therefore not an AI-extraction allowance.

Trim irrelevant content when your workflow permits it, choose concise field descriptions and validate missing values and numeric types. A supported scrape.do Ready API uses its target-specific pricing instead of this character formula. Compare the actual schemas and records accepted by your application; AI extraction and a dedicated parser are different outputs.

Output formats, browser controls and automation

HTML, Markdown and JSON are different contracts

HTML is useful when your existing parser needs tags and attributes. Markdown simplifies text ingestion but removes layout information. ScrapingAnt’s /v2/markdown endpoint returns a JSON wrapper containing the URL and Markdown; its extended API exposes browser metadata alongside content. The AI extractor accepts field descriptions, including nested lists and types, and operates on text rather than inferring absent HTML or CSS layout.

scrape.do’s generic returnJSON browser option requires rendering and exposes browser/network information; it does not automatically turn any product page into a product schema. Its Ready APIs do provide parsed JSON for supported surfaces such as ecommerce, search, maps and travel. Check endpoint pagination, localization, field availability and credit charges. Automatic Amazon routing can also change a response from the HTML expected by an older parser. Treat each response format as an explicit versioned integration contract.

Rendering and browser interactions

ScrapingAnt supports custom js_snippet execution, wait_for_selector, headers, cookies and repeated resource-blocking options. scrape.do offers ordered playWithBrowser interactions such as clicking, filling, scrolling and waiting. Both can help with delayed or interactive content, but waiting longer and executing more actions affects latency. Blocking images or scripts can save work while also removing content your parser needs; verify completeness before optimizing.

Geo-targeting and session behavior

ScrapingAnt’s API documents 24 country values. Its separate standalone residential offering advertises broader country coverage, which should not be copied into an API targeting table. Mobile access is available by request rather than a verified universal self-service option. For scrape.do, postal or ZIP targeting requires residential routing through super=true and a compatible geoCode. Validate returned currency, language and location, rather than trusting a country setting alone.

scrape.do session identifiers run from 0 to 1,000,000 and a session closes after five minutes without a request. A failed request can replace its proxy and session. This is not a permanently dedicated IP. A mobile user-agent string also does not prove that the exit address belongs to a mobile carrier. If a workflow depends on continuity, test it over the full intended sequence.

Async processing and integrations

scrape.do’s async API uses a separate job endpoint and X-Token authentication. It supports polling, webhooks, cancellation and structured plugins for supported targets. The extra pool runs independently of the main API, so Hobby can process ten main calls and three async tasks concurrently, with all consumption deducted from the same credit balance. Retrieve results before each task’s ExpiresAt timestamp and make webhook processing idempotent; temporary results are not a permanent dataset.

ScrapingAnt offers MCP tools for HTML, Markdown and text, Make modules, a community n8n node and a GitHub Action. MCP consumes the same API allowance; it does not make agent web access free. Check node availability in your n8n environment and the Make installation route. These integrations help automate calls, while scheduling, persistent storage and end-to-end job recovery remain your workflow’s responsibilities. If that operational ownership is the main issue, compare ScrapingAnt with Apify before choosing.

Performance: measure useful data, not headline success rates

There is no controlled provider-versus-provider speed or success dataset in this comparison. A numerical provider ranking requires the same target set, settings and validation rules before it can guide this decision. Marketing uptime and success claims measure different things, often with an unspecified request mix. Browser-heavy pages, geography, challenge behavior and your required fields all affect the outcome.

“Successful” does not always mean usable

ScrapingAnt documents zero credit cost for API error responses. scrape.do’s billing reference includes successful 2XX responses and target-site 400, 404 and 410 responses, while distinguishing provider-side validation errors. Concurrency 429, retrieval failure 502 and client-cancel 510 are listed as non-billable. A target response can therefore be billable while containing no record you want. scrape.do may review invalid-content cases for credits or custom rules, but parser rejection is not a universal automatic refund for either provider.

For the hypothetical donut, 1,000 ordinary rendered datacenter responses consume 10,000 ScrapingAnt credits or 5,000 scrape.do credits. Allocated entry-plan costs are $1.90 and $0.58, or $0.002375 and $0.000725 per accepted record when 800 pass validation. If that batch is your only monthly usage, cash cost per accepted record becomes $0.02375 and $0.03625. These are examples of accounting, not measured provider quality or standalone packages.

A practical comparison pilot

Run the pilot across the times and regions that matter to your application. Use bounded retries with backoff and jitter, and separate quota errors from target failures. The MDN explanation of HTTP 429 describes rate-limiting semantics; your client should handle those signals rather than repeatedly submitting the same request. Free plans are useful for validating payloads, but their concurrency and allowance may not represent production capacity.

Which provider fits your workload?

For a visual workflow that a non-developer can configure, neither generic API replaces a task builder. Read ScrapingAnt vs Octoparse for that alternative. For a third managed API option, our ScraperAPI vs scrape.do comparison covers a broader retrieval and pipeline suite.

Implementation, maintenance and data handling

Start with an explicit response contract

Keep API credentials in server-side environment variables or a secret store. Build URL parameters with a URL encoder; nested target URLs contain their own query strings and must not be concatenated carelessly. ScrapingAnt supports x-api-key authentication, while scrape.do’s generic and async routes use different documented authentication patterns. Avoid putting complete credential-bearing URLs into request logs.

Specify browser mode, geography and output format deliberately. Log status, content type, credit usage and a request identifier separately from sensitive content. Confirm whether a JSON response is a wrapper around HTML or Markdown, a browser-network payload or a parsed target schema before accessing fields. Preserve representative fixtures for parser maintenance and add completeness checks before writing records to the database.

Control timeouts and repeat processing

ScrapingAnt documents a default 60-second timeout within a 5–60 second range. scrape.do documents a 5,000–120,000 millisecond parameter and a 60-second default in its overview. Give the client enough time to receive the provider’s timeout response, and distinguish a canceled connection from a completed billable call. For jobs, acknowledge a webhook only after safely recording its event and use a stable task identifier to avoid duplicate inserts.

A useful production design separates retrieval, parsing, validation and storage. Failed retrieval can be retried; an incomplete record may need a different rendering profile or parser fix. Save credits by reprocessing retained content when appropriate instead of refetching the same page for every schema change. Keep independent limits on monthly spend, per-domain load and queue size, then alert before reaching the allowance.

Proxy mode requires a separate security decision

Both providers offer proxy-style integration using the API subscription. Documentation discusses disabled TLS verification, and scrape.do also documents trusting its CA certificate. Use direct HTTPS API calls as the default integration path where practical. If proxy mode is necessary, review trust scope in an isolated client rather than disabling certificate checks across your application. Combining your own browser with provider-managed rendering can also duplicate work without improving output.

Review data permissions and retention

Check access permissions, target terms, personal-data handling and the processing arrangements appropriate to your application. The Robots Exclusion Protocol specification explains how robots rules are expressed; it does not grant permission to collect or reuse data. For private or sensitive inputs, establish what the provider retains, who can access it and how deletion is handled before sending credentials or content. This comparison does not claim an independent security or proxy-sourcing audit.

Strengths and limitations

Support and customer evidence: what to verify

Provider testimonials and review-directory scores can help identify recurring questions, but they do not establish an apples-to-apples satisfaction ranking. This guide does not invent numerical ease-of-use, support or value scores. Review date, sample size, plan and use case matter, and a five-star account may have a very different target mix from your own.

ScrapingAnt’s published support ladder moves from email and documentation to custom snippets, debugging, account management and enterprise onboarding. scrape.do advertises dedicated support at Business and contract-oriented support options higher up. Neither ladder proves a response-time guarantee on every plan. Ask both providers the same concrete question using a minimal authorized URL, expected schema, geography, requested browser settings and sanitized error details.

Evaluate whether support identifies the exact failure, explains its credit treatment and suggests a reproducible fix. For business-critical collection, request the applicable SLA, escalation route, support hours and compensation terms in writing. Separate an engineer helping configure a request from a fully managed extraction service that maintains parsers and deliveries for you.

Final verdict: choose around accepted data and operating effort

ScrapingAnt is a strong starting point for a smaller developer workflow, especially when Markdown, MCP or flexible AI extraction matters. Its low entry bill and larger Free allowance are useful, but expensive residential rendering and AI character charges need careful budgeting. Paid concurrency wording should lead to a measured load test, not an assumption of unlimited throughput.

scrape.do deserves the first cost evaluation for sustained ordinary browser/residential work, supported structured targets and queued batches. Lower mode multipliers can buy more retrieval per credit pool, while Ready APIs and async tools can reduce implementation work. That advantage still depends on accepted output, domain-specific charges, plan concurrency and result retention.

Compare actual monthly cash outlay, cost per accepted record and maintenance time after a pilot. If scheduling, datasets and hosted code execution are the main missing pieces, our existing ScraperAPI vs Apify guide and Octoparse vs scrape.do comparison explain the broader platform tradeoff. A cheaper nominal request is useful only when it produces the data your application can use.

Frequently asked questions