Extract clean markdown content from a web page using a headless browser with full JavaScript rendering. Strips nav/header/footer by default. Returns structured data: title, URL, markdown content, meta tags, timing. Use this instead of WebFetch when you need JS-rendered content or clean markdown.

Low-level page fetch with JS rendering. Returns raw output as string. Supports HTML dump, text dump, link extraction, or arbitrary JS evaluation.

Scrape many URLs in parallel with one call. Each URL is rendered with full JS, then returned as a result object. Far faster than calling kloakt_extract in a loop. Use when you already have a known list of pages to pull.

Discover outbound links on a page. Renders the page, returns its anchors as absolute URLs paired with anchor text, optionally filtered by a query substring and/or restricted to the same domain. Use to find where to go next before extracting, e.g. locate article links on an index page.

Breadth-first crawl from a start URL, returning clean content for each page visited. Budget-limited (max_pages) and depth-limited (max_depth), stays on the start domain by default, with cycle detection. Use to gather a small section of a site in one call. Network-bound — keep max_pages modest.

Get an indexed accessibility/structure tree of a page — an agent-vision substitute when you can't see pixels. Returns compact nodes (index, tag, depth, visible, clickable, role, text, input type/value, href, label). Use to understand page layout and decide what to act on; pass interactive=true to get only the actionable elements (links, buttons, inputs).

Capture a real PNG screenshot of a page. kloakt has no rasterizer, so this delegates to a locally-installed Chrome/Chromium/Edge; it errors clearly if none is found. For structure without pixels, prefer kloakt_snapshot (always available).

Open (or reattach to) a persistent, named browser session. Unlike the stateless tools, a session keeps its cookies AND its page/JS/DOM state alive across separate calls, so you can log in once and keep driving the same page over many turns. A background daemon is auto-started if needed. Use this to begin a multi-step interaction; follow with kloakt_session_act / kloakt_session_read, and kloakt_session_close when done.

Act on a persistent session's page. One tool multiplexes four actions via action: 'navigate' (go to url), 'click' (click a CSS selector), 'type' (set an input's value) or 'eval' (run a JS expression). State persists, so this is how you drive a session opened with kloakt_session_open.

Read the current state of a persistent session's page without changing it. mode='text' returns document.body.innerText; mode='snapshot' returns an indexed accessibility/structure tree (pass interactive=true for only the actionable elements). Use to observe what the page looks like now.

List the names of all open persistent sessions on the daemon.

Close a persistent session, dropping its pages and cookies. Call this when you are done with a session opened by kloakt_session_open to free resources.