Open Browser Control by Sumeet Singh

Open Browser Control lets AI agents running on your own machine see and control your Firefox tabs. It's built for the Model Context Protocol
(MCP) ecosystem — pair it with Claude Desktop, Cursor, Kiro, or any other MCP-compatible client, and your agent gains browser tools: navigate, click, type, take screenshots, read the DOM, switch tabs, hover, execute JavaScript, and more.

Local-only by design. The extension connects to a small bridge server running on your own machine (localhost). It never talks to a remote server and no data leaves your device through the extension.

Session isolation. Each connected agent gets its own named tab group so parallel agent sessions don't step on each other. The sidebar shows an activity log and control-mode switches (AI / User / Collaborative) so you can take over mid-task and hand control back to the agent seamlessly.

Setup:
1. Install this extension.
2. In your MCP client, add the command: npx open-browser-control
3. Open the Open Browser Control sidebar in Firefox — your agent's browser tools are live.

Source, issues, and docs: https://github.com/smankoo/open-browser-control