When developing agents, enabling AI to access websites that require login (such as Zhihu and Xiaohongshu) is a common pain point. Traditional crawlers rely on reverse-engineered APIs and can easily break after platform updates, while also carrying compliance risks. Browser MCP drives a real Chrome browser and provides a local MCP server solution. It uses your logged-in browser state to read and operate on web pages, requiring no Cookies and no API reverse engineering.

Core Capabilities

This plugin mainly provides the following capabilities:

  • Local operation: The MCP server and Chrome extension run only locally, and traffic stays on localhost (127.0.0.1), so data does not leave the machine.
  • Real interaction: Powered by a real Chrome browser, it supports public pages, pages rendered with JavaScript, and pages that require a login state.
  • Automated actions: Supports background actions such as clicking, scrolling, typing, pressing keys, and selecting dropdowns, without switching away from the page you are currently viewing.
  • Data security: It does not request Cookies permissions, does not reverse-engineer APIs, does not bypass CAPTCHAs, and the login state is stored only in the local Chrome Profile.

Installation and Enablement

Install this plugin in a DSH environment using the following command:

dsh plugin --profile web add "github:ywleeo/browser-mcp#main"

After installation is complete, restart dsh web, and the tools will appear in the mcp__browser__* format (such as mcp__browser__browser_read and mcp__browser__browser_snapshot).

Typical Usage

After installation and enabling, you can issue instructions directly using natural language in conversation:

  • Search and extraction: Search Zhihu for answers about MCP, or extract content from Xiaohongshu notes.
  • Media download: Download all images from a Xiaohongshu note to a specified path.
  • Content reading: Read the content and comments of a Douyin video.
  • Background operations: Open a web page, fill in the search box based on a screenshot, and click search.

Notes

  • Localization and compliance: All connection logs and media download statuses are saved locally. The plugin behavior is fully compliant: it does not reverse-engineer internal APIs, does not bypass CAPTCHAs, does not steal Cookies, and does not perform bulk scraping.
  • Permission boundaries: The login state is not leaked, and the extension does not request Cookies. If the AI triggers operations with external effects, such as liking, bookmarking, or publishing, it must ask the user for confirmation before execution.
  • Data integrity: Downloaded media includes SHA-256 checksums. Comment retrieval supports resumption from a checkpoint.