Tools available to your AI agent

Captures the current page state as structured SiFR JSON. Returns salience-scored elements with semantic relationships, computed styles, spatial positions, and actionable selectors. 5-15KB per capture vs 200-400KB raw HTML.

• url — navigate to URL before capturing (optional)
• tabId — capture a specific browser tab (optional)
• browser — target browser: "Chrome", "Firefox" (optional)
• preset — capture detail level: "normal", "minimal" (optional)

Returns the text content of the captured page in a readable format. Title, URL, and extracted text content. Useful for understanding page content without parsing SiFR nodes.

• sessionId — session from a previous capture (required)
• selector — CSS selector to read specific section (optional)

Search for elements in the captured page by text, tag, selector, or salience level. Returns matching nodes with full SiFR detail.

• sessionId — session from a previous capture (required)
• text — search by visible text content (optional)
• tag — filter by HTML tag (optional)
• selector — filter by CSS selector (optional)

Get full details about a specific element — all attributes, computed styles, spatial position, children, parent relationships.

• sessionId — session from a previous capture (required)
• nodeId — element ID from SiFR capture (required)

Perform a single action on the page — click, type, scroll, select, navigate. Returns mutation diff showing what changed.

• sessionId — session from a previous capture (required)
• action — "click", "type", "scroll", "select", "navigate" (required)
• target — element ID or CSS selector (required)
• value — text to type or option to select (for type/select actions)

Perform multiple actions in sequence as a single operation. Optionally submit a form after all steps. Reduces "element changed since capture" errors between clicks.

• sessionId — session from a previous capture (required)
• steps — array of actions [{action, target, value}] (required)
• submit — element to click after all steps (optional)

Scroll within a page and recapture — discover content below the fold without a full page reload.

• sessionId — session from a previous capture (required)
• action — "scroll" (required)
• target — element to scroll within (required)

List all open browser tabs across all connected browsers. Returns tab ID, URL, title, and active status. Works cross-browser — Chrome and Firefox tabs in one response.

• browser — filter by specific browser (optional)