Capterra Scraper - Software Ratings & Pricing

Turn a Capterra product page into a row of structured data. Paste the URLs of the software products you care about and get back the overall rating, the four rating dimensions with the number of reviews behind each one, the published starting price and how it is charged, free-trial and free-version flags, the category, and the vendor's own description.

One row per product. No API key, no login.

What you get per product

Identity — name, product_id, slug, url, logo_url, description, category and breadcrumbs (the category path the product sits under, such as Home › Project Management Software › monday.com).

Ratings — rating and review_count for the headline score, then each dimension separately:

• ease_of_use_rating and ease_of_use_reviews
• customer_support_rating and customer_support_reviews
• value_for_money_rating and value_for_money_reviews
• functionality_rating and functionality_reviews
• recommend_percent

Pricing — starting_price, starting_plan (the plan that price belongs to, such as Pro or Plus), pricing_model (what the price is charged per, for example Per User), payment_frequency (Per Month, Per Year), has_free_trial and has_free_version.

Every dimension carries its own review count, and that matters

This is the detail most comparisons drop. On Slack's profile the overall score rests on 24,194 reviews, but the dimensions are not evenly evidenced:

Dimension	Rating	Reviews behind it
Ease of use	4.6	22,749
Functionality	4.6	22,130
Value for money	4.5	15,307
Customer support	4.4	13,566

A support score built on 13,566 reviews and an ease-of-use score built on 22,749 are not equally settled, and averaging the four dimensions as though they were gives every product a number that is quietly wrong. Both halves are returned so you can weight them yourself, or discard a dimension whose sample is too thin to carry an argument.

Paste URLs, not id numbers

The input takes the address of the product page. Bare id numbers are refused on purpose, and the reason is worth stating plainly: a Capterra id cannot be guessed or inferred. While building this, id 189442 looked like a plausible guess for a well-known design tool. It is a portfolio-management product called Pinnakl. A typed id does not fail, it returns a different company, and nothing in the output tells you that happened. Copying the URL from the page you are actually looking at removes the whole class of mistake.

capterra.com and the country domains such as capterra.co.uk are both accepted, and query strings are ignored.

The mistake this was built around

A Capterra product page does not only describe its product. It also carries the alternatives Capterra recommends alongside it: up to eight products and three separate rating blocks on a single page.

An early version of this scraper took the richest block on the page and returned Slack's ratings under Notion's name. The row looked perfect. It had a plausible score, a large review count, a name, a price. Nothing about it suggested it was describing a different company.

So every field is read from the block that matches the product id in the URL you supplied, and every neighbouring block is ignored. If the page carries no block for the id you asked for, you get an error naming that product rather than a row that quietly belongs to somebody else.

One field was dropped for the same reason instead of shipped. The page carries what reads as a sentiment split, and it is tempting: a clean positive and negative percentage, right there. Measured across six products with ratings from 4.19 to 4.72, it read 85/15 every single time. It is part of the page furniture rather than a measurement of anything, so publishing it would have handed six different products the same confident, meaningless number. It is not in the output.

The ratings come from a block that carries no name of its own, so it is only used once it has been tied to the product you asked for. The structured data at the top of every page states which product the page describes, and its review count is an exact integer; the block is accepted only when the two agree. On a disagreement the rating fields come back empty rather than borrowed.

The alternatives are worth seeing to understand why that matters. They are not just different numbers, they are incoherent ones: monday.com appears on Slack's page with 11,125 ease-of-use reviews and on Notion's page with 22,418, against the same 6,103 total in both. Reading the nearest block gets you a confident figure that belongs to another company and does not even hold together.

Numbers as the site means them

recommend_percent is published as a fraction, 0.9024. Left alone in a column headed "recommend", that invites being read as nine out of ten. It is returned as 90.2, the percentage the site itself displays.

Ratings are published to fourteen decimal places (4.66390427378689) and are rounded to two. Prices arrive with a doubled currency glyph ($$15) which is collapsed to $15. A vendor that publishes no price gets an empty starting_price, never a zero, because a zero in a price column means free.

Run summary

Every run writes a summary record: products requested and returned, how many carried a rating and a price, how many offer a free trial, the total number of reviews the run covers, and how many pages did not come through.

That last count is the one to watch. It is not a failure of the product id and not an empty result, it is a page worth requesting again. Raising Attempts per product is the fix; three is usually enough, and across nineteen products in testing the count was zero.

What people use this for

Competitive tracking. Run a set of competitors on a schedule and keep the rows. Rating movement is slow and mostly noise, but review counts and published prices change in ways that are worth knowing about the week they happen.

Pricing research. starting_price together with pricing_model and payment_frequency is the comparable form of a number that every vendor presents differently on its own site.

Category landscape. Feed in every product in a category and you have the whole field in one table: who has volume behind their score, who has a free tier, who publishes a price at all.

Vendor due diligence. The gap between a product's ease-of-use score and its customer-support score, weighted by how many reviews sit behind each, tells you more about what buying it feels like than the headline rating does.

Scale and cost

Each product is one request. Eight products took 19 seconds end to end in testing, and the Actor runs in 256 MB because there is no browser involved. Set Maximum products to cap a run: it is applied after duplicate URLs are removed, so it is a real ceiling on both time and cost.

A product that cannot be read never becomes a silently missing row. It becomes an entry in the errors record with a code and a message naming the URL, so a run of fifty tells you exactly which two need another attempt.