Most screenshot APIs hand your agent a 500KB base64 image that torches the context window. We return a ~30KB preview, extracted page text, and a signed URL — same information, a fraction of the tokens.
# Screenshot any URL — one POST curl -X POST https://api.pagevistly.com/v1/screenshot \ -H "Authorization: Bearer sk_live_..." \ -d '{ "url": "https://example.com", "response": "agent", "detail": "low", "max_tokens": 4000 }' # → context-friendly JSON { "title": "Example Domain", "text": "This domain is for use...", "preview": "iVBOR...", # ~30KB "est_tokens": 4743, "full_res_url": "https://..." }
What your agent actually receives when it asks to "see" a webpage.
{
"image": "data:image/png;base64,
iVBORw0KGgoAAAANSUhEUg..." [500KB]
}
One image burns most of the context window.
{
"preview": "iVBOR..." [~30KB],
"text": "Pricing. Starter $5...",
"full_res_url": "https://...",
"est_tokens": 1800
}
~90% less context, same information.
We benchmarked stripe.com/pricing with our own API — real renders, real token counts.
A 197KB PNG becomes ~269,000 base64 chars in your context.
91.7% less, same information — and the text is searchable & citable.
Full-page or viewport, any format, cache-hit responses in milliseconds.
Headless Chrome at scale is a part-time job. We already did it — cookie banners, lazy loading, web fonts, retries — so you don't have to.
Official MCP server with max_tokens budgets and detail presets. Your agent gets a ~30KB preview + extracted text instead of a 500KB raw image.
Full-page or viewport, PNG/JPEG/WebP, retina scale, custom device emulation. Cached responses return in milliseconds.
URL or raw HTML to print-perfect PDF. Headers, footers, page ranges — everything headless Chrome can do, minus the DevOps.
Template-based social cards. Pass a title and author, get back a 1200×630 image. Perfect for programmatic content sites.
One flag removes cookie consent walls and ads before capture. Clean pixels, every time — our most-loved option.
API keys with per-key quotas and rate limits. Usage dashboard included. No surprise bills — overage is opt-in.
Drop two lines into Claude Desktop, Cursor, or any MCP-compatible agent. Your agent gains eyes: it can screenshot pages, archive receipts as PDF, and verify what a site actually looks like.
# Inside your agent's tool loop tool_call: read_page url: "https://stripe.com/pricing" detail: "low" # compact preview + text max_tokens: 4000 # → context-friendly response { "title": "Stripe Pricing", "text": "Pricing. Starter $9...", # extracted page text "preview": "iVBOR...", # downscaled, ~30KB "full_res_url": "https://...", # signed, fetch if needed "est_tokens": 1800 }
Start free. Pay only when your agents get busy.