Bridge tool for AI to drive your real Chrome session
chrome-bridge, from Siropkin, is an MCP server plus Chrome extension designed to give AI agents controlled access to a user's active Chrome session for browser-based tasks. It exposes tab listing, full-page HTML/text retrieval, screenshots, and click-and-type automation over a local WebSocket, enabling AI-driven localization and in-page workflows. The tool targets developers and power users who need authenticated, real-session access while keeping all processing local and telemetry-free.
What tasks can you actually use it for?
The tool is aimed at browser-centered automation and analysis, specifically ai-text-localization and similar workflows. It can list open tabs and return the active tab's URL or title, retrieve full page text or HTML for analysis, capture high-fidelity screenshots, and perform click-and-type actions so agents can interact with form fields or navigate pages. Use cases include localized copy extraction, credentialed page inspection, and visual verification.
How reliable are the page captures and automation actions?
chrome-bridge produces outputs tied to the live browser, which preserves logged-in state and extensions; that underlying fact means HTML retrieval and screenshots reflect the real page as the user sees it. The extension's screenshot and full-page retrieval support visual and textual AI analysis, and browser-driven clicks/typing operate on the real DOM. However, dynamic client-side behavior on complex single-page apps may still require human review of automated actions.
What inputs does it accept and what are its limits?
The tool accepts the active Chrome session as its input surface and provides page HTML, text, and images back to the agent. It requires Google Chrome v117+ and Node.js v18+, and it expects clients that speak the Model Context Protocol, so compatibility depends on an MCP-capable agent. It is not a headless replacement and does not create isolated browser instances, which limits use cases where an isolated environment is mandatory.
Is setup approachable and how is privacy handled?
Setup uses a zero-dependency Node CLI and a Chrome extension that binds tovia a local WebSocket, which keeps all processing on the machine. The extension shows a pill overlay that narrates agent commands, offering visible feedback about actions. The developer states there is no telemetry and execution is local, so privacy controls and visible narration help users monitor and limit what the agent can do.
Suitable for developers who need direct, authenticated browser access
chrome-bridge is a practical option for developers and power users who require AI-driven interaction with an existing, authenticated browser session and who prioritize local processing. It demands MCP-capable clients and current Chrome and Node runtimes, and automated results should be validated manually when interacting with dynamic or sensitive pages. Use it as a developer tool to augment, not replace, human verification in web workflows.
