Screenshot for bytes, render_link for a hosted URL you can embed, extract for text, render_artifact for a branded document — each tool's description names the others and says when not to use it, which is unusually helpful in a set this easy to confuse. The watch tools are the part worth setting up carefully: watch_test dry-runs a config without creating a baseline, and the noise controls for ignored regions and text exist because a naive visual diff fires on every rotating banner. rendex_account is read-only and free, so an agent can check its own remaining quota before starting a batch.
A hosted rendering service driven from an agent: screenshots and PDFs, reader-mode extraction, hosted image URLs, branded documents, and scheduled change monitoring — all from one API key and one credit pool.
- A capture of any webpage, raw HTML or Markdown as PNG, JPEG, WebP or PDF, with full-page scrolling, dark-mode emulation, viewport size from 320 to 3840 wide, downscaling, ad blocking, and Mustache templating so one HTML template plus a data object becomes a finished image (rendex_screenshot).
- Reader-mode extraction as Markdown, JSON or HTML from the same Chromium render pass — so it sees content after JavaScript runs, which is what fetch-only readers miss on SPAs — returning the article body with title, byline and excerpt, and options to hide cookie banners or specific selectors first (rendex_extract).
- A signed, hosted, edge-cached image URL rather than the bytes, taking the same parameters as a screenshot plus an expiry between 60 and 2592000 seconds — the shape you want for a dynamic og:image (rendex_render_link).
- A branded artifact in one call: Markdown or HTML turned into a PDF, a PNG and a hosted share page, with a logo, accent colour, font, header, footer and PDF page size, returned as hosted URLs so nothing needs storing on your side (render_artifact).
- Change monitoring on a schedule, with a real-Chrome visual diff and a highlighted overlay, a text diff, or both — created, listed, read, edited in place, run on demand and deleted, with each run recording whether it changed, a diff score, and signed before, after and overlay images (watch_create, watch_list, watch_get, watch_update, watch_run, watch_runs, watch_delete).
- A dry run for a watch before you create one: render the proposed config once and report what was captured and whether the page is reachable, without creating a watch, a baseline or a diff (watch_test).
- Account state from inside the agent — plan, credits used against the monthly limit, when it resets, the per-minute rate limit, and an upgrade link — read-only and costing no credits (rendex_account).
A Rendex account. Locally, `npx -y @copperline/rendex-mcp` with the key in RENDEX_API_KEY; if you put that in a project `.mcp.json`, gitignore it. The hosted endpoint at https://mcp.rendex.dev/mcp needs no install and is fronted by OAuth 2.1 with PKCE, so an OAuth-capable client signs in with a one-time email code and no key at all, each caller billing against their own plan. Free covers 100 calls a month at 3 per minute, rising through Starter and Pro to Enterprise. Geo-targeted capture is Pro and above, and watch check intervals have a per-plan floor.
One command plus a key — npx -y @copperline/rendex-mcp, then supply credentials