anomalyco/browser-control

What it solves

Browser Control allows AI coding agents to interact with a user's actual web browser rather than a separate, headless instance. This enables agents to perform tasks using the user's existing logged-in sessions, browser profiles, and installed extensions, removing the need for the agent to handle complex authentication or session management manually.

How it works

It operates as a driver that connects a local relay to a browser extension. An agent sends Playwright code snippets to the relay, which then executes them via the extension in the user's Chromium-family browser. It includes a "skill" to teach agents the best workflow (e.g., inspecting the page before acting) and can be exposed as tools via an MCP (Model Context Protocol) server.

Who it’s for

It is designed for developers using AI coding agents (like Claude Code or OpenCode) who want those agents to be able to automate tasks in their real browser environment securely and efficiently.

Highlights

• Session Persistence: Uses real browser profiles and logged-in sessions instead of headless browsers.
• Human-in-the-Loop: Includes a handoff() function to pause agent execution for human-only steps like CAPTCHAs or 2FA.
• Tab Adoption: Can attach to and control existing open browser tabs.
• Smart Inspection: Provides snapshot() and ariaSnapshot() helpers to help agents understand page structure using stable references to avoid DOM drift.
• Network Capture: Can record authenticated API exchanges to derive direct API clients without exposing secrets.
• Read-Only Mode: Offers a session type that rejects keyboard and mouse input to prevent accidental mutations.