If you use Claude with Safari on a Mac through Computer Use, you will quickly notice a specific limitation: Claude\'s screenshots only show the above-the-fold viewport. Anything that would require scrolling to see is invisible to a screenshot. This is not a bug. It is a direct consequence of how Computer Use handles Safari, and there is a clean workaround.
Why screenshots are above-the-fold only
Computer Use assigns Safari to "read" tier, which permits screenshots but blocks all forms of interaction including scrolling. A screenshot captures whatever Safari is currently displaying in its window. If Safari has not been scrolled (and Computer Use cannot scroll it), the captured viewport is the initial above-the-fold state.
For pages where the most important content is above the fold (most homepages, most landing pages, most ad creatives), this is fine. For pages where the most important content is below (long articles, product pages with detailed specs, FAQ sections, long-form sales letters), the screenshot misses the substance.
The clean workaround: curl the text
The screenshot is for visual context. The text content lives in the HTML. Curl the URL from the terminal and you get the full text of the page, including everything below the fold:
curl -s -L "https://example.com" | python3 -c "import sys,re; t=sys.stdin.read(); print(re.sub(r'<[^>]+>', ' ', t))"For server-rendered pages (which is the majority of the web outside of single-page apps), this returns the complete textual content. Claude reads it directly, no screenshot needed. The above-the-fold limitation only constrains the visual layer of the workflow.
What to do about JavaScript-rendered pages
Some pages render their content client-side after Safari loads them. React, Vue, Angular, and similar frameworks. For these, curl returns mostly empty HTML and the text-extraction trick does not work. Several options:
Use the Safari Reader View URL
Safari\'s Reader View extracts the main article content from many pages. If Reader View works for a page, the URL with ?reader=1 appended (or the Reader-rendered HTML) sometimes contains the content in cleaner form.
Use a headless browser via the shell
For genuinely client-rendered content, install Playwright or Puppeteer via the terminal and have Claude run a headless Chromium that renders the page and extracts the text. This is heavier but bypasses both the above-the-fold limitation and the curl-cannot-render-JS limitation.
npx -y @browser/cli "https://example.com" --get-textSwitch to Chrome for that specific URL
If the page is heavily interactive and you need full visual + textual context, this is the moment to open Chrome and use Claude in Chrome. The Safari workflow is for the majority of read-and-analyse research; the minority of complex client-rendered pages may need a different tool.
How to know whether the screenshot caught what matters
The fastest sanity check: ask Claude to describe what it sees in the screenshot, then compare against the curl text. If the curl text mentions content (a testimonial, a product spec, a FAQ entry) that Claude\'s description of the screenshot does not, that content is below the fold. Decide whether you need a visual of it specifically, or whether the text alone is enough.
For 80 percent of research tasks, the text alone is enough. For brand audits and conversion-rate optimisation work, the visual matters more and the screenshot limitation can be more painful.
What about resizing Safari to capture more?
One legitimate workaround: resize Safari\'s window before taking the screenshot. A taller Safari window captures more of the page in a single screenshot. Computer Use cannot resize Safari directly (that would require interaction), but you can resize it manually before triggering Claude\'s screenshot.
This is a clunky but real option. If your daily research workflow on Safari benefits from larger screenshots, resize Safari to fill your full vertical screen space and leave it that way. On a 27-inch monitor, a full-height Safari window can show 3,000 to 4,000 pixels vertically, which is several screen-heights of content captured in a single screenshot.
Print preview as a workaround for long-form content
One unusual trick: use Safari\'s print preview (Cmd+P) to view a full-page rendering of any URL. The print preview shows the whole page as it would appear printed, which means all content is visible in a single scrollable view. Computer Use can screenshot the print preview window the same as it would screenshot a normal Safari tab.
The catch: Cmd+P opens the print dialog, which Claude cannot trigger (it would be typing into Safari, which is blocked). You would need to trigger it yourself, then ask Claude to screenshot the resulting preview.
The fundamental tradeoff
The above-the-fold limitation is a deliberate Anthropic product choice, not a technical oversight. Anthropic wants the path to full browser interaction (clicks, scrolling, typing) to go through Claude in Chrome, not through Computer Use on Safari. The limitation is the cost of that design choice.
For research-and-analyse work, the workaround is well-understood: curl for full text, screenshot for above-the-fold visual, and accept that visual content below the fold requires either Chrome, AlliHat, or human assistance.
Bottom line
Above-the-fold-only screenshots are a real limitation in the Safari + Claude workflow. The mitigation is to rely on curl for full-page text and accept that the visual context is partial. For 80 percent of research tasks, this is enough. For the remaining 20 percent that need below-the-fold visual context, switch to Chrome + Claude in Chrome or AlliHat for that specific task.
At ScaleWise VA we use the curl-first pattern as our default and only screenshot when visual context genuinely matters. If you want help building this kind of layered research workflow for your Shopify operations, book a free 30-minute call.
Why this limit matters less for Shopify research than you'd think
For most Shopify operations research, the above-the-fold-only screenshot limit on Safari is genuinely not a constraint, because the work we do is mostly text analysis where curl returns the entire page content regardless of viewport.
Specific examples where the screenshot limit doesn't matter:
- Auditing competitor product page COPY, curl returns every word, including all the below-the-fold sections (reviews, related products, FAQ)
- Extracting competitor brand voice patterns, curl gives the full text of multiple competitor blog posts in seconds
- Reading supplier documentation, almost always server-rendered, curl works perfectly
- Analyzing competitor Klaviyo email content, the email HTML is text-extractable via curl
Where it does matter: brand-feel research where the visual identity is the thing being analyzed. For these tasks (e.g., "does this competitor's homepage feel premium enough that we should reposition?"), we either resize Safari to capture more vertical space, or open the same URL in Chrome with the Claude in Chrome extension which can scroll natively.
For ongoing operations work specifically, this limit costs about 5-10 minutes per week of workaround time. Not enough to invest in AlliHat or switch browsers. The shell workflow + occasional Chrome handoff covers it.
If you'd like to see this kind of operational workflow built into your Shopify retainer, book a discovery call.