How Chrome 146’s New MCP Feature Disrupts AI Browser Automation
Chrome 146 adds native MCP support that lets AI agents control the currently logged‑in browser session, eliminating the need for headless mode or credential transfer, expanding automation possibilities while introducing new security considerations.
Chrome 146 has been released and includes native support for the Chrome DevTools Protocol (MCP). By enabling a toggle in chrome://inspect/#remote-debugging, an AI agent can directly control the browser session that is already running, instead of launching a new instance.
Previously, using AI to operate a browser required either a headless mode— which is easy to detect— or the cumbersome migration of login cookies and tokens. Both approaches were fragile, and some developers resorted to relay plugins that proved unstable.
Developer Petr Baudis demonstrated a concrete use case: he asked Claude to clean up LinkedIn connection requests that were trying to sell something. Claude opened Petr’s already‑logged‑in LinkedIn page, analyzed each invitation, and batch‑ignored them, all without any additional authentication steps.
No re‑login needed . The agent reuses the user’s existing login state.
No fingerprint detection . Because the agent drives a real browser, it avoids the detection mechanisms that target automation tools.
Broader automation scope . Tasks such as auto‑filling forms, interacting with government websites, or testing custom web applications become feasible with a lower barrier.
However, the ability for an agent to manipulate a real browser session raises control‑and‑audit concerns; clear permission boundaries and operation logs are required to mitigate security risks.
The current Chrome MCP client can become unstable when handling several hundred tabs. To improve the experience, Petr created a skill called chrome-cdp-skill (available at https://github.com/pasky/chrome-cdp-skill). It can be installed with:
npx skills add https://github.com/pasky/chrome-cdp-skillAlternatively, simply enabling the debugging toggle in chrome://inspect activates the capability. The upcoming OpenClaw release will also support this feature, and token consumption for browser‑automation tasks is expected to drop significantly.
Using a browser remains one of the most important capabilities for large language models, and Chrome’s openness to real‑browser automation is a boon for many existing workflows.
AI Engineering
Focused on cutting‑edge product and technology information and practical experience sharing in the AI field (large models, MLOps/LLMOps, AI application development, AI infrastructure).
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.