CurlShot — Screenshot-as-a-Service API for Developers and AI Agents

What is CurlShot?

CurlShot is a screenshot-as-a-service API aimed at developers and AI agents. As the site puts it: "Paste a website address and get a screenshot of the whole page as an image or PDF. Free to try, nothing to install." The service accepts a URL — or raw HTML and markdown — and returns a rendered capture in png, jpeg, webp, pdf, mp4, webm, or gif, either as the file itself or as a JSON response containing a hosted file link.

The intended audience is engineering workflows rather than end users: documentation snapshots, website monitoring, PDF generation, QA, and agentic web browsing. Notably, the documentation is explicitly written "for AI agents" and exposes an MCP server at https://curlshot.com/api/mcp with take_screenshot and list_devices tools, plus an OpenAPI 3.1 spec and docs in markdown and llms.txt form. No company name, country, or launch history is published on the site.

How CurlShot Works

The workflow is a three-line integration. First, register at curlshot.com/register to receive a free access key. Second, send a GET request (or POST with JSON for complex options) containing the key and the target URL — the documented example is curl "https://curlshot.com/api/v1/screenshot?access_key=YOUR_ACCESS_KEY&url=https://example.com&format=jpeg". Third, receive the rendered file directly, or poll GET /api/v1/jobs/{id} for asynchronous and bulk jobs; GET /api/v1/usage reports remaining quota.

Authentication accepts the key as a query parameter, an X-Access-Key header, or a Bearer token. Capture behavior is controlled with parameters: full page or element selectors, clip regions, named device viewports such as iphone_15_pro or desktop_full_hd, and cleanup options including ad, tracker, cookie-banner, and chat-widget blocking, custom styles and scripts, forced dark mode, and hiding chosen selectors. Setting cache=true makes repeated captures of the same page free — cache hits do not consume quota.

Main Features

• Multi-format capture — png, jpeg, webp, pdf, mp4, webm, and gif, with optional JSON responses carrying hosted file links.
• Flexible targeting — full-page, element-selector, and clip-region captures across named device viewports (iphone_15_pro, pixel_9, macbook_air_13, desktop_full_hd, and more).
• Page cleanup — block ads, trackers, cookie banners, and chat widgets; inject custom styles and scripts; force dark mode; hide selectors.
• Free caching — cache=true makes repeated captures free, so monitoring loops do not burn quota.
• Async and bulk rendering — background jobs with webhooks, and a bulk endpoint accepting up to 100 renders per call.
• MCP server for AI agents — take_screenshot and list_devices tools at /api/mcp, with OpenAPI 3.1 and llms.txt documentation.

Pricing

CurlShot uses a freemium quota model. The Free plan offers 100 screenshots per month with 10 requests per minute and 1 concurrent render. Starter costs $19 per month for 3,000 screenshots (60 req/min, 5 concurrent), Growth costs $79 for 15,000 (150 req/min, 15 concurrent), and Scale costs $249 for 60,000 (400 req/min, 40 concurrent). One successful render equals one screenshot; failures and cache hits are free. Exceeding quota returns HTTP 402 with an upgrade_url. Rate limits return 429 with a Retry-After header, and plan-gated options return 403 feature_not_available.

Common Use Cases

• Documentation snapshots — archiving how pages looked at a point in time.
• Website monitoring — scheduled captures with caching to keep costs near zero.
• PDF generation — converting reports or invoices rendered as web pages into PDFs.
• QA and visual checks — element-level captures across device viewports.
• Agentic browsing — letting AI agents capture pages via the MCP server during research tasks.

Limitations

The service blocks private and local addresses (localhost, 10.x, 192.168.x) with a host_not_allowed error, so it cannot screenshot internal systems. Some options are plan-gated and return 403 on lower tiers. Rate limits and concurrency are enforced per plan. Beyond the product surface, there is no published company name, country, launch year, or SDK story — integrators work with raw HTTP. And because it is a hosted renderer, pages behind logins or bot walls may not capture as expected, though the site does not detail anti-bot workarounds.

Summary

CurlShot is a tightly scoped, well-documented screenshot API: one HTTP call turns a URL or raw HTML into an image, PDF, or video capture, with developer-grade options (device viewports, element selectors, cleanup, caching), automation primitives (async jobs, webhooks, bulk), and a genuinely agent-friendly surface (MCP server, OpenAPI, llms.txt). Pricing is transparent and the free tier is enough for real evaluation. The gaps are corporate opacity — no company or SLA information — and plan-gated features. For developers and agent builders who need programmatic page captures, it is a credible, low-friction option.