MCP 服务器

PagePixels Screenshots MCP Server

Enables AI assistants to capture website screenshots, automate browser interactions, and manage recurring screenshot configurations across 150+ global locations. It also supports AI-powered domain research and visual change monitoring for any web page.

README

pagepixels-screenshots-mcp-server

A Model Context Protocol (MCP) server for the PagePixels Screenshot API. Enables AI assistants like Claude, Cursor, Windsurf, and other MCP-compatible clients to capture screenshots, manage screenshot configurations, monitor page changes, and perform AI-powered domain research — all through natural language.

Features

Instant Screenshots — Capture any URL or render raw HTML to an image on demand
Multi-Step Browser Actions — Click, type, submit forms, navigate pages, and wait for elements before capturing
Scheduled Screenshots — Create recurring capture configurations on minute, hour, day, week, month, or year intervals
Geo-Located Captures — Screenshot from 150+ real locations worldwide (countries, US states, and major cities)
Change Notifications — Monitor pages for visual changes with Slack and webhook alerts
Domain Research — AI-powered structured data extraction across multiple domains
Full Configuration Management — Create, read, update, delete, and list screenshot configurations programmatically
OAuth 2.1 Authentication — Secure, spec-compliant MCP auth via Streamable HTTP transport

Requirements

A PagePixels account (free tier available — no payment info required)

Quick Start

Claude.ai

Open Settings (bottom-left) and navigate to Integrations
Select Add custom connector
Enter the server URL:
```
https://mcp.pagepixels.com/mcp
```
Complete the OAuth authorization when prompted in the browser

Once connected, PagePixels tools will be available in all new conversations.

Cursor / Windsurf / Claude Desktop / Other MCP Clients

Connect over Streamable HTTP by pointing your client at:

https://mcp.pagepixels.com/mcp

For clients that support remote MCP servers with OAuth, authentication is handled via the standard MCP OAuth 2.1 flow with PKCE — authorize through the browser prompt when first connecting.

Tools

Screenshot Capture

Tool	Description
`snap`	Capture a screenshot of any URL with full control over viewport, format, quality, wait conditions, JS/CSS injection, multi-step browser actions, and geo-location
`snap_html`	Render raw HTML content into a screenshot with the same capture options as `snap`
`capture_screenshot`	Trigger an immediate capture for an existing screenshot configuration

Screenshot Configuration Management

Tool	Description
`create_screenshot_config`	Create a new configuration with optional scheduling (minutes through years)
`get_screenshot_config`	Retrieve a configuration by ID
`update_screenshot_config`	Update any property of an existing configuration
`delete_screenshot_config`	Delete a configuration and all its captured screenshots
`list_screenshot_configs`	List all configurations in the account with pagination and time filters

Screenshot History

Tool	Description
`list_config_screenshots`	List all screenshots for a specific configuration
`list_all_screenshots`	List all screenshots across the account

Change Notifications

Tool	Description
`list_config_change_notifications`	List change notifications for a specific configuration
`list_all_change_notifications`	List all change notifications across the account

Domain Research

Tool	Description
`create_domain_research`	Submit an AI-powered data extraction job across one or more domains with custom field definitions
`get_domain_research_status`	Check the status of a domain research job
`get_domain_research_report`	Download completed research results as JSON or CSV
`list_domain_research_reports`	List all domain research reports in the account

Account & Utilities

Tool	Description
`get_account_limits`	Get current usage and account limits
`get_job_status`	Check if a screenshot job has completed
`list_real_locations`	List all 150+ supported geo-locations for screenshot capture

Usage Examples

Once connected, interact naturally with your AI assistant:

Take a screenshot:

"Take a screenshot of https://example.com in PNG format at 1440px wide"

Multi-step capture:

"Go to https://example.com, click the login button, fill in the email field with test@example.com, submit the form, then take a full-page screenshot"

Scheduled monitoring:

"Create a screenshot configuration for https://example.com/pricing that captures every 6 hours as JPEG"

Geo-located capture:

"Screenshot https://example.com from Japan with a Retina scale factor"

Domain research:

"Research these 5 competitor domains and extract their pricing tiers, founding year, and team size"

Account overview:

"Show me my current usage and remaining credits"

Capture Options

The snap and snap_html tools support extensive configuration:

Viewport: page_width, page_height, scale_factor (1x or 2x Retina)
Output: image_format (jpeg, png, webp), quality (1–100 for JPEG), thumbnails
Full Page: fullpage, fullpage_advanced, incremental_scroll for lazy-loaded content
Timing: wait (ms delay before capture), wait_for (CSS selector to appear)
Injection: js_inject, css_inject for custom page modifications
Privacy: no_ads, no_cookie_banners, no_tracking
Scripting: disable_js, disable_third_party_js
Geo-Location: real_location (150+ locations), latitude/longitude, time_zone
Auth & Headers: cookies, headers, user_agent
Element Capture: selectors to capture a specific DOM element
HTML Extraction: html_only to return fully rendered HTML instead of an image
Caching: ttl to control cache duration (0 for always fresh)

Multi-Step Actions

Automate browser interactions before capturing with multi_step_actions:

[
  { "type": "click", "selector": "#login-btn" },
  { "type": "text_field", "selector": "#email", "value": "user@example.com" },
  { "type": "text_field", "selector": "#password", "value": "secret" },
  { "type": "submit" },
  { "type": "wait_for_selector", "selector": ".dashboard" }
]

Supported action types: click, hover, change, redirect, javascript, evaluateJs, css, text_field, select, checkbox, submit, wait, wait_for_selector

Authentication

Authentication is handled through the standard MCP OAuth 2.1 flow with PKCE. When connecting through Claude.ai, Cursor, or other OAuth-capable MCP clients, authorize through the browser prompt when first connecting. No manual API key configuration is needed.

For programmatic access outside of MCP, see the PagePixels API documentation.