网页截图 MCP 服务

PagePixels Screenshots MCP Server

A Model Context Protocol (MCP) server for the PagePixels Screenshot API. Enables AI assistants to capture instant and scheduled screenshots, generate images from custom HTML, take screenshots after performing browser actions, and fetch the full raw HTML from dynamic websites using 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
• Real Location Screenshots — Screenshot from 150+ geographical locations (countries, US states, and major cities)
• AI Analysis Screenshots — Capture screenshots and have AI analyze the images, either with a custom text prompt or by extracting structured data fields into JSON
• Custom HTML Screenshots — Create screenshots from HTML and data available in the MCP client
• Multiple Image AI Analysis — Have AI compare up to 5 images and 5 prompts
• Extract HTML — Enable AI to get HTML content from dynamic websites that generate their HTML using JavaScript and find CSS Selectors
• 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

MCP Server URL

Connect over Streamable HTTP by pointing your client at:

https://mcp.pagepixels.com/mcp

Example: Claude.ai Web

1. Open Customize (toolbox icon).
2. Select Connectors.
3. Click Add. Select Add custom connector option.
4. Enter the server URL:
   复制

   https://mcp.pagepixels.com/mcp
   

5. Complete the OAuth authorization when prompted in the browser.

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

Tools

Capture Instant Screenshots

Scheduled Screenshots

Screenshot History

Real Locations

Multiple Image AI Analysis

Extract HTML

Change Notifications

Account

Usage Examples

Once connected, interact naturally with your AI assistant:

Take a screenshot:

"Take a screenshot of https://example.com in PNG format at 540px wide and remove cookie banners"

Multi-step capture:

"Go to https://example.com, fill in the search field with 'text to search for', submit the form, then take a full-page screenshot"

Scheduled monitoring:

"Capture a screenshot of https://example.com/pricing every 6 hours"

Geo-located capture:

"Screenshot https://example.com from Japan and wait 6000ms before taking the screenshot"

AI analysis (simple prompt):

"Take a screenshot of https://example.com and use AI to summarize what the page is about"

AI structured data:

"Screenshot https://example.com/product and extract structured data: product_name (string), price (number), in_stock (boolean)"

Scheduled structured monitoring:

"Every hour, screenshot https://example.com/status and extract a structured status field (enum: ok, degraded, down) plus a list of any visible error messages"

Fetch HTML content:

"Extract the HTML content from https://example.com with PagePixels and check for spelling and grammatical errors"

Account overview:

"Show me my current usage and remaining credits for PagePixels"

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, pdf), quality (1–100 for JPEG), thumbnails
• Full Page: fullpage, fullpage_advanced, incremental_scroll for lazy-loaded content
• Element Capture: selectors to screenshot a specific DOM element
• Timing: wait (ms delay before capture), wait_for (CSS selector to appear)
• Custom Styles: css_inject
• Scripting: js_inject
• Remove/disable: no_ads, no_cookie_banners, no_tracking
• Geo-Location: real_location (150+ locations), time_zone, accept_language
• AI Analysis: analyze_image_with_ai and ai_prompt to analyze a screenshot guided by a custom text prompt, or ai_prompt_mode: "structured" with ai_structured_fields (and optional ai_structured_instructions) to extract typed fields (string, number, boolean, enum, multi) into a JSON object
• Auth & headers: cookies, headers, user_agent
• Caching: ttl to control cache duration (0 for always fresh)

See full list of screenshot options.

Multi-Step Actions

Automate browser interactions before capturing with multi_step_actions:

[
  { "type": "wait_for_selector", "selector": "#email" }
  { "type": "text_field", "selector": "#email", "value": "user@example.com" },
  { "type": "text_field", "selector": "#password", "value": "secret" },
  { "type": "submit" },
  { "type": "wait", "value": "8000" },
  { "type": "click", "selector": ".chart-filter" }
]

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

See Multi-Step section of MCP Help Guide

Authentication

Authentication is handled through the standard MCP OAuth 2.1 flow with PKCE and uses DCR (Dynamic Client Registration). When connecting through Claude.ai, ChatGPT, VS Code, 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.

Transport

This server uses the MCP Streamable HTTP transport, hosted at:

https://mcp.pagepixels.com/mcp

All MCP clients that support remote Streamable HTTP servers can connect directly. OAuth 2.1 with PKCE is enforced on all connections.

Resources

• Screenshot MCP Server Help Guide — Guide to using the PagePixels MCP Server
• PagePixels Documentation — Full API reference
• Model Context Protocol — MCP specification

Support

Contact PagePixels anytime by email:

support-tickets@pagepixels.com

Or contact us through our Support Form.

License

MIT