Reading Page Content

import { Aside } from ‘@astrojs/starlight/components’;

Browser Automation requires an active Pro license.

ptah_browser_content

Return the rendered content of the current page. Use this when an agent needs to reason about the DOM or extract text.

Arguments

Name	Type	Required	Default	Description
format	enum	no	text	One of html, text, markdown
selector	string	no	—	Limit the extraction to matches of this selector
maxChars	number	no	100000	Truncate output to this many characters

Formats

Format	Description
html	Full rendered HTML including inline scripts and styles
text	Readable plain text extracted via the Readability algorithm
markdown	Readable content converted to Markdown, preserving headings and links

Example

{
  "tool": "ptah_browser_content",
  "arguments": {
    "format": "markdown",
    "selector": "main"
  }
}

`markdown` is the best format to feed back into an LLM — it preserves structure while staying compact.

ptah_browser_evaluate

Run arbitrary JavaScript in the page’s context and return the serialised result.

Arguments

Name	Type	Required	Description
script	string	yes	JavaScript source; the last expression or return value is returned
args	any[]	no	Optional arguments passed to the function
timeoutMs	number	no	Defaults to 5000

Example

{
  "tool": "ptah_browser_evaluate",
  "arguments": {
    "script": "return document.title;"
  }
}

You can also pass arguments:

{
  "tool": "ptah_browser_evaluate",
  "arguments": {
    "script": "(selector) => document.querySelectorAll(selector).length",
    "args": ["a[href]"]
  }
}

Constraints

• The script runs in the page, not in Node. Browser globals (window, document, fetch) are available; Node globals (require, process) are not.
• Return values must be JSON-serialisable. DOM nodes and functions cannot be returned directly.
• Scripts that throw return an error with the original stack trace.