If your agent already drives a browser, use that for pages behind a login or a multi-step flow. For public URLs this is the better engineering: every capture runs the same pipeline, can be taken from a named country with matching locale and time zone, and costs a fraction of a cent rather than a browser session plus vision tokens.
An MCP server for the Site-Shot screenshot API. It renders the page in real Chromium and returns the image to the client, with ad and cookie-banner removal on by default — which produces both a cleaner picture and fewer vision tokens when a model looks at it. Country proxies let you see the page as a visitor elsewhere would.
- capture_screenshot — a page as PNG or JPEG at a viewport or device size you choose, with full_page, block_ads, block_cookie_banners, wait_ms and max_height as options, and country, language, time_zone and geolocation for locale-accurate rendering
- capture_full_page — the same capture with the whole scrollable height enabled, as a convenience wrapper
A Site-Shot API key in SITESHOT_API_KEY. Clients run npx -y site-shot-mcp over stdio. Country must be a two-letter ISO 3166-1 alpha-2 code — the server rejects a full country name before spending a render, and strict_country (on by default) errors when the country has no proxy instead of quietly falling back to the US. Pass width, height and wait_ms explicitly whenever the size matters: omitted values are decided by the API and can change.
One command plus a key — npx -y site-shot-mcp, then supply credentials