jumodada/Drissionpage-MCP-Server

What it solves

DrissionPage MCP Server allows AI models (like Claude or Codex) to control a web browser with high precision. It bridges the gap between a model's reasoning and actual browser execution, enabling the AI to navigate websites, extract data, fill forms, and interact with complex visual elements that traditional scrapers or simple automation tools might struggle with.

How it works

It acts as a local Model Context Protocol (MCP) server that exposes 69 typed tools based on the DrissionPage library. These tools allow a model to:

• Interact via DOM or Coordinates: Use standard CSS/XPath selectors for structured data or precise X/Y coordinates for visual-only surfaces (like canvas or maps).
• Natural Motion: Simulate human-like pointer movement using a deterministic 24-step eased cubic path to avoid robotic, instant jumps.
• State Management: Handle multiple tabs, manage cookies/storage, and observe network traffic.
• Verification: Take screenshots and accessibility snapshots to verify the result of an action before proceeding.

Who it’s for

• AI Agent Developers: Those building agents that need to perform real-world web tasks.
• QA Engineers: For creating automated, AI-driven browser testing workflows.
• Researchers: For technical research involving web automation and accessibility.
• Power Users: Anyone using MCP-compatible clients (like Claude Desktop or Codex) who wants their AI to have a fully functional browser.

Highlights

• 69 Atomic Tools: Comprehensive coverage of navigation, tab management, element interaction, and network observation.
• Deterministic Natural Motion: Reproducible human-like mouse movements for more natural interaction.
• Vision-Ready: Designed to work with multimodal models that can identify coordinates from screenshots.
• Deep Browser Access: Supports Shadow DOM, iframes, and direct Chromium/CDP control.
• Type-Safe: Built with Pydantic validation to ensure the AI sends correctly formatted commands.