browser

Automate and control a running Chrome browser via the Chrome DevTools Protocol.

Updated Sep 2, 2026
One-click install
npx skills add https://github.com/nbiish/tts-mcp --skill browser-nbiish
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: browser
Source: https://github.com/nbiish/tts-mcp/tree/main/.agents/skills/browser-harness
Command: npx skills add https://github.com/nbiish/tts-mcp --skill browser-nbiish

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Automating real web pages often requires launching a separate browser instance, losing the user's logged-in sessions, and fighting brittle selector-based automation. This Skill connects directly to the user's already-running Chrome over CDP, so agents can navigate, click, scrape, and verify pages using the user's actual browser state. ## Core Features & Use Cases - Direct CDP control: Navigate pages, click by screenshot-derived coordinates, run JavaScript, and issue raw CDP commands against the user's live Chrome. - Screenshot-driven interaction: Capture screenshots to locate targets, click at pixel coordinates that pass through iframes, shadow DOM, and cross-origin frames, then re-screenshot to verify. - Remote and parallel browsers: Start isolated cloud browsers via the Browser Use API for parallel sub-agents or headless servers, with optional cloud profiles and proxies. - Use Case: Scrape hundreds of static pages with bulk HTTP requests, or automate a multi-step workflow on a site where the user is already logged in, without launching a new browser. ## Quick Start Ask the agent to open a specific URL in a new tab of your running Chrome and report back the page information.

Frequently Asked Questions about browser

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate Chrome without launching a new browser?▼

Connect to the user's already-running Chrome over the Chrome DevTools Protocol instead of launching a new instance. The browser-harness CLI auto-starts a daemon that attaches to the existing browser, preserving logged-in sessions and open tabs.

How do I click elements inside iframes or shadow DOM?▼

Take a screenshot, read the target's pixel coordinates, and click at those coordinates. Input.dispatchMouseEvent is handled at the compositor level, so clicks pass through iframes, shadow DOM, and cross-origin frames without selector work.

Can I run parallel browser sessions for sub-agents?▼

Yes, start a remote cloud browser per sub-agent using start_remote_daemon with a distinct BU_NAME for isolation. This requires the BROWSER_USE_API_KEY environment variable and supports cloud profiles, proxies, and custom timeouts.

Why does my browser automation session go stale?▼

Default daemon sessions can go stale, and omnibox popups appear as fake page targets. Call ensure_real_tab() to re-attach to a real page; the daemon also auto-recovers from stale sessions on the next call.

When should I use selectors instead of coordinate clicks?▼

Drop to DOM selectors only when the target has no visible geometry, such as hidden inputs or zero-size nodes. For everything visible, screenshot plus coordinate clicking is the default and avoids brittle selector hunting.