A category page with 40 products on it, twenty pages deep, is 800 rows. Octoparse's free plan lets you export 50,000 rows a month, so that job fits sixty-two times over. Then the counter stops. Each of those runs also has to happen on your own machine, with the app open, two at a time. No cloud, no schedule, no API. That is the actual shape of the free deal, and it is both more generous and more fiddly than the marketing implies.
This review was rebuilt against Octoparse's own pricing page, release notes, help centre, API reference and GitHub organisation on 13 August 2026. Where the vendor's pages contradict each other, both numbers appear below, because they do contradict each other in three places.
What the free plan actually gives you
The pricing page lists the free tier as 10 tasks, 1 device, 1 user, and 50,000 rows of exported data per month, with a hard ceiling of 10,000 rows on any single export. Extraction is local, two concurrent runs, and the app keeps the last 5 runs of history. Exports go to Excel, CSV, JSON, HTML and XML. Cloud extraction, task scheduling, IP rotation and API access are all switched off.
Pages per run are not capped. The earlier version of this review had that backwards. It claimed the days of a free plan with unlimited pages were over. They are not. Octoparse's own blog post of 18 May 2026 describes the free tier as "10 tasks, unlimited pages/run". What gets metered is rows leaving the tool, not pages entering it. A task that walks 4,000 listing pages and returns 300 deduplicated rows costs you 300 rows of quota. This changes how you build: filter and deduplicate inside the workflow, before export, not afterwards in a spreadsheet.
The 10,000-row export ceiling is the limit you meet first. A single mid-sized marketplace scrape blows through it, and the fix is to split the task by category or date range and export in batches. Annoying rather than fatal, and worth knowing before you build one giant task and discover you cannot get the data out of it in one piece.
Templates are partly free. Octoparse states that 74.9% of its 600-plus template library is available on the free plan. That decimal point is the vendor's, not ours.
The template count nobody can agree on
Four Octoparse pages give four different figures for the same library. The pricing page says "500+ preset scraping templates". The May 2026 press release says 600+. The templates gallery says only "hundreds". The MCP page says 200+ are reachable from an AI client. None of these is a lie, they are counting different things, but any competing review that hands you one number with confidence copied it from somewhere without checking. Templates also matter less than the count suggests: what you want is a template for your target, and the gallery is the only way to find out.
The desktop app, August 2026
Current version is 10.1.1, released 28 July 2026. The download page asks for Windows 10 64-bit or higher. The help centre's installation article still says Windows 7 64-bit or higher, which is documentation drift rather than a supported configuration. The Mac build wants macOS 10.14 Mojave or higher and is listed as x64, so on Apple Silicon it runs through Rosetta.
The version number jumped. Octoparse shipped 8.9.0 on 2 March 2026 and 10.0.0 on 8 May 2026, with no 9.x in between. If a comparison table tells you Octoparse is "version 8", it was written before May 2026 and everything else in it is that old too.
Three things changed in the 10.x line that matter to anyone who scrapes logged-in pages:
- 10.0.0, 8 May 2026. Chrome Mode reworked to carry login sessions and to survive CAPTCHA-protected pages.
- 10.1.0, 1 July 2026. Local browser scraping, which runs the task against a browser you are already signed into. Auto-detect learned to see inside shadow DOM and nested iframes.
- 10.1.1, 28 July 2026. Sign-in with GitHub, plus fixes.
The shadow DOM change is the quiet one. Sites built out of web components hide their content inside shadow roots, where an ordinary XPath expression cannot reach. Before this release, auto-detect would look at such a page and see nothing worth extracting, which read as "the tool is broken" rather than "the DOM is nested". If you gave up on a target in 2025 for that reason, it is worth another ten minutes.
How a task gets built
The build loop is four steps, and the first two carry the product.
- Enter a URL. The page opens in Octoparse's embedded browser.
- Auto-detect, or click. Auto-detect looks for lists, tables and pagination and proposes a scraper. When it guesses right on a conventional listing page, you have a working task in two clicks and no further thought. When it guesses wrong, and on anything with an unusual layout it will, you click each field by hand and the tool starts to feel like work.
- Build the flow. Actions such as "click next page", "loop through each item" or "enter this search term" appear as an editable diagram.
- Run and export. Locally or in the cloud, out to a file, a database or Google Sheets.
Because the tool drives a real browser, it handles what breaks naive extractors: logins, AJAX content that arrives after the page settles, infinite scroll, "load more" buttons, cascading dropdowns, pagination with no visible Next link, and values hidden in attributes rather than text. That list has been true of Octoparse for years and is still the honest reason to pick it over a copy-paste extension.
When clicking is not enough
Point-and-click picks elements by generated paths, and generated paths break on any layout the generator did not anticipate. Octoparse exposes manual XPath and regular expression editors for exactly those cases, and the expressions are ordinary ones. Our primers on XPath for scraping and regex extraction apply unchanged inside the tool.
Data cleaning is built in: trim whitespace, find and replace, match with a regex, add a prefix or suffix, reformat dates, decode HTML entities. Version 8.7 added conversion of relative timestamps such as "1 month ago" into real dates, which is the single most common cleanup job on review and forum data.
One habit saves more time than any of this. Before building a visual workflow, open the browser network panel and check whether the page is fed by a JSON endpoint. If it is, you want API-style extraction rather than a click flow, and no visual builder will be faster or more stable than the endpoint itself.
The MCP server, added May 2026
The largest change since this review was first written has nothing to do with the desktop app. On 12 May 2026, in a PR Newswire release from Octopus Data Inc. of Irvine, California, Octoparse announced support for the Model Context Protocol. The same release claims the company is "serving over 6 million users globally", which is a vendor number with no independent audit behind it.
In practice it means an AI client can drive Octoparse directly. The official server is MIT-licensed and exposes six tools: search_templates, execute_task, export_data, search_tasks, start_or_stop_task and redeem_coupon_code. It authenticates with OAuth or an API key, and Octoparse lists ChatGPT, Claude, Cursor and Gemini CLI among the supported clients. During launch the quota is up to 2,000 records a week at no cost, which is a separate allowance from the desktop plan.
Two caveats. The repository carries a single commit on its main branch, so this is a published artefact rather than a project developed in the open, and its issue history will tell you nothing about how fast bugs get fixed. And nothing on the page says what the weekly quota becomes after launch pricing ends. Neither is a reason to avoid it, but treat the free 2,000 records as an introductory rate.
The same GitHub organisation holds octoparse-cli and an agent-skills repository, both touched in 2026. What it does not hold is a maintained client library. The octoparse package on PyPI is a third-party client by Tarique Anwer, last released as 1.6.0 on 4 March 2021. If you plan to talk to the API from Python, you are writing the HTTP calls yourself.
Pricing, read off the vendor's pages on 13 August 2026
| Plan | Monthly | Billed annually | Tasks | Cloud concurrency | API |
|---|---|---|---|---|---|
| Free | $0 | $0 | 10 | none, local only, 2 runs | none |
| Standard | $69 | $58.14/mo | 100 | up to 3 concurrent cloud processes | data export only |
| Professional | $249 | $209.16/mo | 250 | up to 20 concurrent cloud processes | advanced, includes run control |
| Enterprise | custom | custom | 750 | custom | advanced |
Annual billing is advertised as a 16% saving. Enterprise customers can buy unlimited task storage as a $299 add-on.
Then there are the meters. The subscription is not the whole bill, and this is the part most reviews of Octoparse leave out entirely:
- Residential proxy traffic: $3 per GB. Listed on the pricing page and repeated in the help centre. IP rotation is described as a plan feature, and the bandwidth it consumes is charged separately. If you have never sized proxy traffic before, our note on rotating residential proxies covers where the gigabytes actually go, and the short answer is images and fonts you never wanted.
- CAPTCHA solving: $1 per 1,000, or $1.50 per 1,000 for Cloudflare challenges. The help centre adds one sentence that deserves its own paragraph: "Failed attempts still consume credits."
- Pay-per-result templates: $0.001 to $3 per thousand results. A three-thousand-fold range, which means the price of a template job is not knowable until you pick the template.
That CAPTCHA line is the one to think about. A target that challenges half your requests and defeats the solver half the time still bills you for every attempt. On a 10,000-page crawl with a 50% challenge rate, you are paying for 5,000 solves whether or not they worked, and at Cloudflare rates that is $7.50 for a job that may have returned nothing. Small money on its own, useless money in aggregate, and a signal you are pointed at the wrong target.
Where Octoparse contradicts Octoparse
Three disagreements between the vendor's own pages, all live on 13 August 2026:
- The pricing page says Standard is $69 a month and Professional $249. The help centre article on plan-based versus usage-based charges says $75 and $208.
- The pricing page gives Standard up to 3 concurrent cloud processes. The help centre article on cloud extraction says Standard users "can run up to 6 concurrent tasks in the cloud (up to 6 cloud nodes available)".
- Template counts, as above.
Before you buy on the strength of a concurrency number, check what your account page shows after signup. We would tell you which figure is correct, but the only way to know is to hold the subscription, and this review is not affiliated with Octoparse.
What breaks when you scale it up
A visual scraper that works beautifully on one page has failure modes that only appear at volume. These are the ones with numbers attached.
Tasks are a storage quota, not a run quota. The 10, 100 and 250 figures count saved task definitions, not executions. Hit the ceiling and the help centre offers exactly two remedies: delete tasks you no longer need, or upgrade. Anyone monitoring many small targets, one task per competitor, will hit the task ceiling long before any row limit.
Concurrency is the real throughput cap. Three concurrent cloud processes on Standard is the number that decides whether an overnight job finishes overnight. Splittable tasks are divided across nodes, so a well-built task uses that concurrency better than a badly built one, but three is three.
API access does not mean what the plan table implies. The API reference is explicit: token endpoints are open to everyone, task search and data retrieval need Standard, and task execution, parameter updates and all cloud extraction control are Professional and Enterprise only. If your plan is to trigger runs from your own system, that is a $249 a month feature, not a $69 one. Rate limit across the API is 20 requests per second, and going over returns 429.
Automatic export is narrower than the older feature lists say. Octoparse's help centre states that auto-export supports MySQL, SQL Server, PostgreSQL and Google Sheets. Not Oracle, which the previous version of this review listed. The same article carries a constraint that undercuts the word "unattended": the Auto-Export Tool "must remain open" for exports to continue. Google Sheets is the exception, since version 8.7 removed that requirement for Sheets only. Export to Google Drive, Amazon S3 and Dropbox is "only available for Professional and above plan users".
Deletion is final. From the help centre, verbatim: "once any task or data gets deleted, they are removed permanently from our server and would not be retrievable." No trash, no grace period, no undo.
How long cloud data is kept is not documented. The pricing table has a row for data retention, but the published pages we could read do not fill in the values, and no help centre article states a figure. Treat cloud storage as a staging area and export on a schedule. Where the retention boundary sits is not knowable from outside the product.
Where Octoparse stops working
Anti-bot defence moved under it. On 1 July 2025 Cloudflare announced that new domains would default to blocking AI crawlers, calling the day Content Independence Day. That policy targets model training rather than a marketer pulling 500 product prices, but the enforcement machinery does not distinguish, and it sits in front of a large share of the commercial web. The 2025 Thales/Imperva Bad Bot Report put automated traffic at 53% of everything it observed, roughly three quarters of that classed as malicious. Sites are tuned for that ratio, and a visual scraper announcing itself with default headers gets classified accordingly.
Octoparse's answer is metered proxies and metered CAPTCHA solving, which works on ordinary defences and gets expensive on serious ones. Where a site fingerprints the browser, scores behaviour, or issues managed challenges on every request, you reach the point where a managed extraction service or a hand-built stack with proper fingerprint control is cheaper than the credits you are burning.
The legal boundary is closer than the tool suggests. Nothing in a point-and-click interface warns you that the target's terms of service exist. The case people cite as permission, hiQ Labs against LinkedIn, did not end the way the headlines suggested: hiQ won on the Computer Fraud and Abuse Act question and then lost on breach of contract, settling in 2022 with a $500,000 judgment and a permanent injunction against scraping LinkedIn. Public does not mean unrestricted. Personal data carries its own rules regardless of where you found it, which our note on GDPR and web scraping goes through.
Millions of rows. Extraction at that scale, across many domains, is a pipeline problem rather than a tool problem. You need retries, deduplication, schema validation, alerting when a site changes its markup, and somewhere to put the output. Octoparse gives you the extraction and roughly none of the rest.
The alternatives, with prices checked the same day
| Tool | Entry price | Free tier | What it is best at |
|---|---|---|---|
| Web Scraper | $50/mo (5,000 URL credits) | free browser extension, local runs | quick jobs in the browser, cheap cloud scheduling |
| Browse AI | $48/mo (2,000 credits) | 50 credits/mo, 90-day retention | monitoring a page and alerting on change |
| Apify | $29/mo plus usage | $5 of platform credit | ready-made actors, developer control |
| Diffbot | $299/mo (250,000 credits) | 10,000 credits/mo at 5 req/min | extraction as an API, no interface to learn |
| ParseHub | from $189/mo | up to 5 projects | the closest desktop equivalent, at nearly three times the price |
Two of those entries correct the earlier version of this review directly.
ParseHub is not the cheap comparable it used to be. Its own help centre puts paid plans at $189 a month against Octoparse Standard's $69, which is a different market rather than a competing tier. Its marketing pages also return nothing but a page title to a plain HTTP fetch, so everything checkable about ParseHub now comes from its Zendesk help centre. That is not evidence the product is dead. It is evidence you cannot verify its current plans without an account.
Browser extensions are stronger than the summary suggested. The Web Scraper extension has 800,000 users, 4.1 out of 5 from 1.1K ratings, and shipped version 1.111.13 on 15 July 2026. It is maintained, it is free, it runs locally, and for a one-off list it will beat installing a desktop application. Its cloud tiers are $50 a month for 5,000 URL credits with 2 concurrent scrapers and 30-day retention, $100 for 20,000, and from $200 for unlimited credits with 60-day retention and residential proxy at $2.50 per GB. Our roundup of scraper extensions covers the rest of that shelf.
Bright Data, Zyte and similar platforms sit above all of this, and Import.io has moved toward managed data delivery rather than a desktop builder. Those are procurement decisions, not tool choices.
The rule of thumb still holds, with the prices attached. Extensions for a quick grab. Octoparse for repeatable no-code tasks at $69 a month. Apify or a proxy platform when you need scale. Code when the target fights back.
Who it fits
Octoparse earns its place with one specific person: an analyst who needs the same twenty sources refreshed on a schedule, has no engineering support, and is scraping targets that do not aggressively defend themselves. For that person the auto-detect plus template path is genuinely fast, the cloud tier turns a manual chore into a feed, and $69 a month is cheaper than half a day of anyone's time.
It fits less well if you are scraping five targets that all sit behind Cloudflare. Or if your volumes run to millions of rows. Or if the data has to arrive clean and on time in a production system, because the failure mode of a visual workflow is silent: the layout changes, the selector matches nothing, the task reports success, and the column is empty until someone notices.
Where it earns its price, and where it does not
Octoparse in August 2026 is a maintained, actively developed product. Five releases between 8 May and 28 July 2026, a working MCP server, real improvements to login handling and DOM traversal. The free tier is a working tool rather than a demo, and the paid entry price is low for the category. Those are facts off the vendor's own release notes, not impressions.
What you should price in before subscribing: metered proxy and CAPTCHA credits on top of the subscription, run control gated behind the $249 tier, an undocumented retention window, and an auto-export tool that has to stay open for anything except Google Sheets. None of those appear in the headline comparison, and together they decide whether the tool costs $69 a month or several times that.
Where it runs out of road, on stubborn anti-bot systems, volume in the millions, or data feeding a production system on a deadline, a hand-built scraper or a finished dataset serves better. Receiving structured output on a schedule is what our data extraction service and ongoing data as a service exist to do, and the trade is control over the process against not having to maintain it. If you are earlier than that and still working out which category of tool you need, our overview of what web scraping is sets out the ground.