Headless Browser

Headless Browser

Use the web_browse tool to open web pages, run JavaScript, click and type into elements, and extract rendered content. Ideal for SPAs and JS-heavy sites where web_fetch returns little or no content.

When to use

• User asks to open a specific URL or "go to" a page
• Site content is loaded by JavaScript and web_fetch returns empty or minimal HTML
• User wants to interact with a page (click buttons, fill forms, scroll)
• User needs content that only appears after JS runs (e.g. infinite scroll, modals)

Tool: web_browse

Single tool with an action parameter. Maintains one browser session; use close when done to free resources.

Actions

• navigate — Load a URL. Params: url (required), optional wait_ms (default 2000) for JS to settle.
• get_content — Get current page text as markdown. Optional max_chars (default 50000).
• get_html — Get raw HTML. Optional max_chars.
• click — Click element by CSS selector. Param: selector.
• type — Type text into element. Params: selector, text.
• js — Run arbitrary JavaScript. Param: script. Use for complex interactions.
• scroll — Scroll page. Params: direction (down/up), optional amount (pixels, default 500).
• back / forward — Browser history.
• close — Close the session. Call when finished to release the browser.

Workflow

1. navigate to the URL (use higher wait_ms for heavy SPAs).
2. Optionally scroll to load lazy content, then get_content or get_html.
3. To interact: click or type with a CSS selector, then get_content if the page changed.
4. When done, call close.

Tips

• Prefer get_content over get_html for summaries; use get_html only when structure matters.
• If content is missing, increase wait_ms on navigate or scroll first.
• Use simple, robust selectors (e.g. button.primary, #submit) for click/type.
• Close the session when finished to avoid leaving the browser in use.