{
  "id": 6062190,
  "title": "chrome-bridge: let any AI agent drive your real logged-in Chrome",
  "url": "https://urgent.news/2026/09/06/chrome-bridge-let-any-ai-agent-drive-your-real-logged-in-chrome",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-06T23:16:22.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/siropkin/chrome-bridge-let-any-ai-agent-drive-your-real-logged-in-chrome-b5n"
  },
  "original_language": "en",
  "account": "chrome-bridge enables any AI agent to control a logged-in Chrome browser, bypassing the need for a clean browser with zero cookies. Unlike other MCP browser bridges, chrome-bridge does not require a configured server or a logged-out session. Instead, it operates by driving the Chrome instance that the user is already logged into.\n\nA small, unpacked extension communicates with a local Node server through a WebSocket, allowing any agent capable of running shell commands to interact with the browser. Examples include running a compact accessibility snapshot with element references using `node cli.mjs snap localhost:8082`, taking a screenshot-free accessibility snapshot, or performing actions like clicking or filling forms using specific element references.\n\nThe use of snapshots, represented as compact text trees with references, proves to be significantly more economical in terms of tokens compared to traditional screenshots. For instance, a snapshot of Hacker News on the front page would display the navigation menu, links, and article titles in a condensed format. This approach allows agents to efficiently communicate with the browser without the overhead of loading full webpages.\n\nImportant features include the persistence of element references across multiple snaps, enabling agents to perform diff comparisons to check for changes without requiring a complete page re-read. Additionally, equipped tabs display a visual indicator and emit live command notifications in the terminal, providing real-time feedback. The extension supports multiple Chrome profiles and can interface with various AI agents, including Claude Code, Cursor, Qwen, GLM, Kimi, or any script interacting with the local HTTP endpoint, without necessitating MCP setup.\n\nThe software is released under an MIT license, has no dependencies, and requires Node.js version 18 or later. The source code and detailed usage instructions can be accessed at the provided GitHub repository link. Feedback is encouraged, particularly on the token cost of snapshots when compared to their practical applications.",
  "summary": "Ask an agent to check something on a website and most tools spin up a clean browser with zero cookies. Playwright and friends drive a browser they launched, with a profile of their own. MCP browser bridges need an MCP-capable client and a configured server. Either way you start every session logged out. chrome-bridge takes the other road: it drives the Chrome you are already logged into. A tiny…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}