Website Screenshot + Markdown for AI Agents

Capture pixel-perfect, clean full-page website screenshots at scale. Built from the ground up for AI agents, MCP servers, visual regression testing, and automated page archiving. It automatically eliminates intrusive cookie banners, triggers lazy-loaded images, handles mobile and desktop viewports, and provides instantly accessible public image URLs ready for multimodal LLMs.

---

What does it do?

This Actor automates browser rendering via headless Chromium to take reliable, high-fidelity screenshots of public web pages. It handles common rendering roadblocks such as cookie consent banners, incomplete lazy loading, and dynamic layouts without requiring custom browser scripting.

Key Use Cases

• AI Agents & MCP Workflows: Provide multimodal language models (Claude, GPT, Gemini and other vision models) with clean visual snapshots of live web pages for visual web browsing, UI analysis, layout comprehension, and multimodal extraction.
• Visual Monitoring & Defacement Detection: Monitor landing pages, pricing tables, and user interfaces on a schedule to detect unwanted layout regressions, broken styling, unauthorized changes, or downtime.
• Archiving & Visual QA: Archive timestamped visual proof of legal disclosures, ad landing pages, or audit responsive web designs across desktop and mobile screen resolutions in batch.

---

Why this one?

Standard headless screenshot tools often fail in real-world scenarios, producing images obscured by cookie dialogs, broken with half-loaded images, or difficult to pipe into automated pipelines. This actor is specifically tailored to solve these problems:

• Cookie Banners & Consent Popups Removed: Automatically dismisses common GDPR/CCPA consent modals using smart DOM button clicking and injects CSS rules to suppress fixed overlays that hide page content.
• Lazy-Load & Dynamic Scrolling: Executes a controlled auto-scroll sequence before capturing, ensuring lazy-loaded images, charts, and dynamic content render properly instead of appearing blank.
• Desktop & Mobile Emulation: Supports crisp desktop captures (1366×768 default) and mobile emulation (iPhone 13 profile), as well as custom viewport dimensions from small mobile screens up to 4K displays.
• High-Throughput Batch Processing: Pass between 1 and 1,000 URLs in a single run with configurable concurrency to maximize throughput without overloading resources.
• Markdown + Metadata (included at no extra cost): The same run also returns the page's main content as clean Markdown (navigation, footer, scripts and other chrome removed, relative links made absolute) plus metadata (title, description, canonical, lang, ogImage, wordCount, linksCount). Ideal for LLM context and RAG, with no second request.
• LLM-Ready Output: Automatically hosts images in Apify's Key-Value Store and outputs public image URLs (screenshotUrl) directly in the dataset, allowing multimodal LLMs to consume the image immediately without complex decoding.

---

How much does it cost?

This actor uses a transparent Pay-Per-Event pricing model:

• Price: $5 per 1,000 screenshots (charged per successful capture).
• Markdown and metadata are included: Page content as Markdown and page metadata come with every successful screenshot at no extra cost; there is no separate event.
• Pay only for success: You are charged only when a screenshot is successfully captured and saved (status: "ok"). If a URL fails due to a DNS error, network timeout, or invalid target, you are not charged for that capture.
• Predictable budgeting: Set spend limits on your Apify account to ensure your runs never exceed your intended budget.

---

Input

Configure the actor using the table below. All settings are passed via the standard JSON input object:

Field	Type	Default	Description
urls	Array<string>	["https://apify.com"]	Required. List of 1–1,000 URLs to capture. Protocol is optional (defaults to https://). Invalid URLs will generate an error record without breaking the run.
device	string (desktop | mobile)	"desktop"	Preconfigured device profile. desktop uses 1366×768; mobile emulates an iPhone 13.
viewportWidth	integer	null	Custom viewport width in pixels (320–3840). When provided, overrides the device profile.
viewportHeight	integer	null	Custom viewport height in pixels (240–2160). When provided, overrides the device profile.
fullPage	boolean	true	When true, captures the full scrollable page height. When false, captures only the initial viewport.
format	string (png | jpeg)	"png"	Image format. Use png for lossless text/charts or jpeg for smaller file sizes.
quality	integer	80	Image compression quality (1–100). Applies only when format is "jpeg".
removeCookieBanners	boolean	true	Automatically attempts to click accept buttons and hides sticky/modal consent overlays via CSS.
scrollToBottom	boolean	true	Automatically scrolls down the page before capture to trigger lazy-loaded media (capped at 10 seconds or 20 screens).
hideSelectors	Array<string>	[]	List of custom CSS selectors to hide before taking the screenshot (applied as display: none !important;).
waitUntil	string	"load"	Navigation wait condition: "load", "domcontentloaded", or "networkidle".
delayMs	integer	500	Additional pause (0–10,000 ms) after the page is ready before triggering the capture.
timeoutSecs	integer	30	Navigation and rendering timeout per URL in seconds (5–120s).
maxConcurrency	integer	3	Number of web pages rendered concurrently (1–5).

Example Input

{
  "urls": [
    "https://apify.com",
    "https://news.ycombinator.com"
  ],
  "device": "desktop",
  "fullPage": true,
  "format": "png",
  "removeCookieBanners": true,
  "scrollToBottom": true,
  "delayMs": 500,
  "timeoutSecs": 30
}

---

Output

Every run outputs items to both the Dataset and the Key-Value Store.

1. Dataset Item Format

For each URL processed, a structured item is pushed to the default dataset:

{
  "url": "https://apify.com",
  "finalUrl": "https://apify.com/",
  "status": "ok",
  "screenshotUrl": "https://api.apify.com/v2/key-value-stores/a1b2c3d4/records/screenshot-c4ca4238a0b9-0",
  "key": "screenshot-c4ca4238a0b9-0",
  "format": "png",
  "width": 1366,
  "height": 4210,
  "bytes": 812345,
  "title": "Apify: Full-stack web scraping and data extraction platform",
  "device": "desktop",
  "truncated": false,
  "markdown": "# Apify

Full-stack web scraping and data extraction platform...

[Pricing](https://apify.com/pricing)",
  "markdownTruncated": false,
  "metadata": {
    "title": "Apify: Full-stack web scraping and data extraction platform",
    "description": "Cloud platform for web scraping, browser automation, and data for AI.",
    "canonical": "https://apify.com/",
    "lang": "en",
    "ogImage": "https://apify.com/og-image.png",
    "wordCount": 1250,
    "linksCount": 84
  },
  "error": null,
  "takenAt": "2026-10-29T22:45:00.000Z"
}

If a URL fails, an error item is recorded, allowing your batch processing to continue seamlessly:

{
  "url": "https://invalid-domain-does-not-exist.xyz",
  "finalUrl": null,
  "status": "error",
  "screenshotUrl": null,
  "key": null,
  "format": "png",
  "width": null,
  "height": null,
  "bytes": null,
  "title": null,
  "device": "desktop",
  "truncated": false,
  "markdown": null,
  "markdownTruncated": false,
  "metadata": null,
  "error": "net::ERR_NAME_NOT_RESOLVED",
  "takenAt": "2026-10-29T22:45:02.000Z"
}

2. Key-Value Store

Binary image files are stored directly in the default Key-Value Store using their unique key identifier with the proper MIME type (image/png or image/jpeg). The screenshotUrl provides a direct, accessible HTTP link to download or view the image.

---

Use with AI Agents / MCP

This actor is purpose-built to integrate into autonomous agent architectures via the Model Context Protocol (MCP) or direct API calls.

1. Using the Apify MCP Server

When using the @apify/mcp-server integration in environments like Claude Desktop, Cursor, or LangChain agents, add the actor to your MCP toolset. The agent can invoke the actor whenever visual verification or page inspection is needed:

{
  "name": "apify/website-screenshot",
  "arguments": {
    "urls": ["https://example.com"],
    "fullPage": true,
    "removeCookieBanners": true
  }
}

The agent receives the dataset containing the screenshotUrl, which can be forwarded directly to multimodal vision models for prompt reasoning.

2. Calling via Apify REST API

You can run the actor synchronously or asynchronously from any script:

curl -X POST "https://api.apify.com/v2/acts/<ACTOR_ID>/run-sync-get-dataset-items?token=<YOUR_APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{
    "urls": ["https://apify.com"],
    "fullPage": true,
    "format": "png"
  }'

3. Calling via Python SDK

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")

run = client.actor("<ACTOR_ID>").call(run_input={
    "urls": ["https://apify.com"],
    "fullPage": True,
    "removeCookieBanners": True,
    "scrollToBottom": True
})

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    if item["status"] == "ok":
        print(f"Captured {item['url']} -> {item['screenshotUrl']}")

---

FAQ

1. How are cookie popups and consent banners dismissed?

The actor uses a multi-layered approach: it scans the DOM for commonly used consent dialog buttons (e.g., "Accept All", "Agree", "Allow Cookies") and clicks them. Concurrently, it injects custom CSS rules to hide prominent sticky overlays and backdrop dialogs that often obscure page content.

2. What happens if a URL fails to load?

If a single URL fails due to a network error, DNS resolution failure, or timeout, it is marked as status: "error" in the output dataset with an informative error message. The actor continues processing the remaining URLs in your batch, and you are not charged for the failed item.

3. Are the screenshots directly accessible by multimodal LLMs?

Yes. Each successful screenshot record includes a public screenshotUrl hosted in Apify's Key-Value Store. Multimodal vision models (Claude, GPT, Gemini and others) can fetch and inspect the visual layout directly using this URL.

4. Is using this actor legal and compliant?

Yes. Capturing screenshots of publicly available web pages is a standard and legitimate web practice. However, users are solely responsible for the URLs they submit and must comply with the target websites' terms of service, copyright considerations, and applicable privacy regulations. This actor does not bypass paywalls or access private data.

5. Can I capture full-length landing pages with heavy lazy loading?

Yes. Setting scrollToBottom: true directs the browser to perform an automated scrolling pass down the page (capped at 10 seconds or 20 screen heights) before snapping the picture. This triggers intersection observers and JavaScript lazy-loading scripts so images and components render completely.

---

Limitations

• Heavy Anti-Bot Defenses: Websites protected by strict anti-bot systems, Cloudflare "Under Attack" mode, or interactive CAPTCHAs (such as Cloudflare Turnstile or reCAPTCHA) may block headless Chromium requests.
• Authentication & Login Portals: The actor is designed for publicly accessible pages. Pages requiring user login, single sign-on (SSO), or session cookies cannot be accessed without custom session credentials.
• Infinite Scrolling Restraints: Pages with infinite scrolling (such as social media feeds) will auto-scroll up to a safe threshold of 20 viewports or 10 seconds to avoid memory overflow and runaway run times.