Web Browsing

Native Apple WebKit browser with 42 actions — navigate, interact, extract content, fill forms, scrape, and automate web pages.

Yukkai includes a built-in web browser powered by native Apple WebKit — the same engine as Safari. With 42 actions covering navigation, interaction, content extraction, tab management, sessions, anti-detection, media handling, and pre-built scraping recipes, it can handle everything from simple page reads to complex multi-page automation flows.

Why WebKit?

Native & lightweight: ~50 MB RAM per tab. No Electron, no dependencies.
Real Safari User-Agent: Natural anti-bot fingerprint — sites see a genuine Safari browser.
Full JavaScript support: SPAs, React, Vue, and dynamic content all work perfectly.
Cookies persist across navigations within a session.

Actions Reference

Navigation

Action	Description
`navigate`	Load a URL. Returns an indexed DOM tree. Handles redirects. Timeout: 60s.
`back`	Go back in browser history.
`forward`	Go forward in browser history.
`wait`	Wait for the page to stabilize (timeout in ms).

Interaction

Action	Description
`click`	Click an element by its DOM index.
`input_text`	Type text into an element by index (clears the field first).
`select_option`	Select a dropdown option by index and value.
`send_keys`	Send keyboard keys (Enter, Escape, Tab, Backspace, or text).
`scroll`	Scroll the page (direction: up/down/left/right, pages: N).

Content Extraction

Action	Description
`extract`	Get the full page content as clean, structured Markdown.
`search_page`	Find text matches on the current page (fast, free).
`screenshot`	Capture the page as a PNG (saved to `~/YukkaiDocs/`).
`evaluate`	Run arbitrary JavaScript and return the result.

Tab Management

Action	Description
`get_tabs`	List all open tabs with their IDs.
`new_tab`	Create a new tab (optionally with a URL).
`switch_tab`	Switch to a tab by its `tab_id`.
`close_tab`	Close a tab by its `tab_id`.

Cookies & Window

Action	Description
`cookies`	Get session cookies (optionally filtered by domain).
`show`	Open a visible browser window the user can see and interact with.
`hide`	Close the visible browser window.

Sessions (Persistent Login)

Action	Description
`save_session`	Save all cookies to disk (`session_name` required).
`load_session`	Restore previously saved cookies (`session_name` required).
`list_sessions`	List all saved sessions.
`delete_session`	Delete a saved session (`session_name` required).
`clear_data`	Clear all cookies and cache.

Advanced Wait Actions

Action	Description
`wait_for_selector`	Wait for a CSS selector to appear (params: `selector`, `timeout`).
`wait_for_text`	Wait for specific text to appear on the page (params: `text`, `timeout`).
`wait_for_network_idle`	Wait for all fetch/XHR requests to complete (param: `timeout`).
`wait_for_dom_change`	Wait for DOM content to change (params: `selector`, `timeout`).

Infinite Scroll

Action	Description
`scroll_and_collect`	Scroll the page and collect elements. Params: `selector`, `max_items` (default 100), `scroll_pause` (default 1500ms), `extract_js` (custom extraction code).

Human-Like Behavior (Anti-Detection)

Action	Description
`human_click`	Natural click with mouse movement and delays.
`human_type`	Character-by-character typing with natural timing.
`human_scroll`	Organic scroll with variable speed.

Media

Action	Description
`collect_media`	List all images/videos on the page (optional URL filter).
`download_media`	Download a specific media URL (params: `url`, optional `filename`).
`download_all_media`	Download all media matching a filter (params: `filter`, `max_items_media` default 50).

Data Export

Action	Description
`export_data`	Export data to a file (params: `data` as JSON string, `format`: json/csv/markdown, `filename`).

Request Interceptor

Action	Description
`start_interceptor`	Start capturing fetch/XHR requests (param: `url_pattern` to filter).
`get_intercepted`	Retrieve captured requests (param: `clear: true` to reset).
`stop_interceptor`	Stop capturing.

Scraping Recipes

Action	Description
`recipe`	Run a pre-built scraping recipe (params: `recipe_name`, `target`).
`list_recipes`	List available recipes.

Available recipes: google_search, github_repo, github_trending, hackernews, reddit, twitter, youtube_search, wikipedia, amazon_search, linkedin_company.

DOM Tree Format

When you navigate to a page, Yukkai returns an indexed DOM tree (Browser Use format):

[0]<nav />
    [1]<a href="/" />
        Home
    [2]<a href="/about" />
        About
[3]<main />
    [4]<input type=search placeholder="Search..." />
    [5]<button />
        Go

Only interactive elements get an [index]. Use this index to click or input text. For example, to fill the search field above and click the button:

web_browse(action: "input_text", index: 4, text: "AI agents")
web_browse(action: "click", index: 5)

Examples

Read a Web Page

web_browse(action: "navigate", url: "https://example.com")
web_browse(action: "extract")

Fill a Search Form

web_browse(action: "navigate", url: "https://google.com")
web_browse(action: "input_text", index: 4, text: "AI agents")
web_browse(action: "click", index: 5)
web_browse(action: "extract")

Scrape with a Recipe

web_browse(action: "recipe", recipe_name: "hackernews", target: "frontpage")

Save & Restore a Login Session

// First time: log in manually or via automation, then save
web_browse(action: "save_session", session_name: "github")

// Later: restore and continue
web_browse(action: "load_session", session_name: "github")
web_browse(action: "navigate", url: "https://github.com/dashboard")

Open a Visible Browser for the User

web_browse(action: "show", url: "https://apple.com")

Scroll and Collect Items

web_browse(action: "scroll_and_collect", selector: ".article-card", max_items: 50, scroll_pause: 2000)

Tips

Prefer extract over screenshot for reading content — Markdown output is structured and searchable.
Use wait_for_selector after clicking buttons that trigger dynamic content, before trying to extract results.
Save sessions for sites you visit frequently to avoid re-authenticating each time.
Use evaluate to run custom JavaScript when the standard actions don't cover your needs.
For mass scraping (10+ profiles or pages), consider the apify tool instead — it handles proxies and rate limiting automatically.