Browser Control MCP
以下摘录自该仓库的 README(片段) · 在 GitHub 读完整 README →
An MCP server paired with a Firefox browser extension that provides AI assistants with access to tab management, browsing history, and webpage text content.
Features
The MCP server supports the following tools:
- Open or close tabs
- Get the list of opened tabs
- Create tab groups with name and color
- Reorder opened tabs
- Read and search the browser's history
- Read a webpage's text content and links (requires user consent)
- Find and highlight text in a browser tab (requires user consent)
- Capture a screenshot of a tab's visible area (requires user consent per tab)
Example use-cases:
Tab management
- *"Close all non-work-related tabs in my browser."*
- *"Group all development related tabs in my browser into a new group called 'Development'."*
- *"Rearrange tabs in my browser in an order that makes sense."*
- *"Close all tabs in my browser that haven't been accessed within the past 24 hours"*
Browser history search
- *"Help me find an article in my browser history about the Milford track in NZ."*
- *"Open all the articles about AI that I visited during the last week, up to 10 articles, avoid duplications."*
Browsing and research
- *"Open hackernews in my browser, then open the top story, read it, also read the comments. Do the comments agree with the story?"*
- *"In my browser, use Google Scholar to search for papers about L-theanine in the last 3 years. Open the 3 most cited papers. Read them and summarize them for me."*
- *"Use Google search in my browser to look for flower shops. Open the 10 most relevant results. Show me a table of each flower shop with location and opening hours."*
Comparison to web automation MCP servers
The MCP server and Firefox extension combo is designed to be more secure than web automation MCP servers, enabling safer use with the user's personal browser.
* It does not support web page modification, page interactions, or arbitrary scripting.
* Reading webpage content requires the user's explicit consent in the browser for each domain. This is enforced at the extension's manifest level.
* Screenshots require the user to authorize each individual tab by clicking the extension's toolbar button. The extension relies on the activeTab permission for this, so the authorization is enforced by Firefox itself and expires when the tab navigates or closes. The extension never requests broad host permissions for screenshots.
* It uses a local-only connection with a shared secret between the MCP server and extension.
* No remote data collection or tracking.
* It provides an extension-side audit log for tool calls and tool enable/disable configuration.
* The extension includes no runtime third-party dependencies.
**Important note**: Browser Control MCP is still experimental. Use at your own risk. You should practice caution as with any other MCP server and authorize/monitor tool calls carefully.
Installation
Option 1: Install the Firefox and Claude Desktop extensions
The Firefox extension / add-on is available on addons.mozilla.org. You can also download and open the latest pre-built version from this GitHub repository: browser-control-mcp-1.5.0.xpi. Complete the installation based on the instructions in the "Manage extension" page, which will open automatically after installation.