Agent Browser

agent-browser

This skill exposes a headless Chromium browser as the browser_session MCP tool.

When to use

• Fetching content from JS-rendered pages (SPAs, React/Vue apps)
• Interacting with a page before reading its content (login forms, search boxes, navigation)
• Observing API calls a page makes (XHR/fetch requests, WebSocket handshakes)
• Taking visual screenshots of a page state

Key properties

• Fully isolated: Each call gets a fresh browser context. No cookies, localStorage, or session state carries over between calls.
• Structured output: Returns title, finalUrl, dom, screenshot (base64 PNG), and networkLog in a single JSON response.
• Declarative steps: Provide an ordered steps array — no arbitrary code execution.

Example usage

{
  "url": "https://example.com",
  "steps": [
    { "type": "wait_for", "selector": "#main-content" },
    { "type": "read_dom", "selector": "article" },
    { "type": "screenshot" }
  ],
  "captureResponseBodies": false
}

Step types

| Type | Required | Optional | Description | |------|----------|----------|-------------| | navigate | url | — | Navigate to a new URL | | click | selector | — | Click an element | | type | selector, text | — | Type text into a field | | wait_for | selector | timeout (ms) | Wait for element to appear | | wait_ms | ms | — | Fixed delay | | read_dom | — | selector | Extract text (defaults to body) | | screenshot | — | — | Capture viewport as base64 PNG |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Vorwo
• Source: Vorwo/agent-browser-skill
• License: MIT

Install and usage instructions live in the source repository linked above.