Browser Connector

Browser Connector lets an AI agent you run on your own machine drive a browser you are already signed into. It is a Chrome extension plus a local MCP server, and it exists for one loop: build, test, debug in a real browser without copying state between tools. The agent can read a controlled tab — its accessibility tree, text, screenshots, console messages and network metadata — and act in it: navigate, click, type, scroll, fill forms, upload and download files. A cursor marks the point being clicked, and screenshots come back as images. Everything stays on your computer. There is no account, no server, and no telemetry: page content travels over Chrome native messaging to the agent host you installed, and nowhere else. The extension needs a companion program, installed separately. It is open source under Apache-2.0 at https://github.com/benedictjeremiahh/browser-connector, which is also where the MCP server and the install steps live. This is a development tool. It does not ask before acting: any paired agent can read and change any site the browser can reach, including where you are signed in. The side panel shows what is connected and every call that was made, and the Control Lease can be revoked at any time.