Skip to content
TakoAPI
Back to skills

midscene-computer-browser

by quanru

Web & Frontend Development
305Downloads reported by clawskills.sh
1Views
0 likesSign in to like this skill
—Not yet rated
Your rating

Brief

Vision-driven browser automation using Midscene.

Description

What This Skill Does

Vision-driven browser automation that operates from screenshots alone, without needing DOM access or accessibility labels. Launches headless Chrome via Puppeteer and interacts with any visible page element using natural language instructions. Works across any frontend technology stack.

Works from screenshots alone, so it handles any frontend stack without needing DOM access, custom selectors, or accessibility attributes.

When to Use It

  • Scraping product prices from e-commerce pages
  • Filling and submitting multi-step online forms
  • Taking screenshots of web pages for review or archiving
  • Verifying UI behavior after a deployment
  • Extracting data from paginated tables or dashboards

Example Workflow

INPUT

User asks: Get the titles of the top 5 articles on Hacker News

AGENT
  1. 1

    Connect to https://news.ycombinator.com

  2. 2

    Take a screenshot to confirm the page has loaded

  3. 3

    Act: read the titles of the top 5 articles from the page

  4. 4

    Close the browser

OUTPUT

List of the top 5 Hacker News article titles extracted from the page

Requirements

  • MIDSCENE_MODEL_API_KEY environment variable (API key from a supported vision model provider)
  • MIDSCENE_MODEL_NAME environment variable
  • MIDSCENE_MODEL_BASE_URL environment variable
  • MIDSCENE_MODEL_FAMILY environment variable

Installation

clawhub install quanru/midscene-computer-browser

Run the CLI command, or paste the prompt into Claude Code, Cursor, Codex or whichever coding agent you use.

Links