Tools

These tools are registered by the current source code under src/tools.

Tool	Category	Use it for	Output shape
browse_and_act	Agent	Run a complex browser task from a start URL using a natural-language instruction.	Legacy plain text result or error text.
browse_and_act_structured	Agent	Run the same agent flow when the client expects structured data.	{ ok, result, instruction, start_url } or { ok: false, error, error_type }.
get_page_snapshot	Observation	Get the active page URL, title, and visible body text.	{ ok, url, title, text, text_length, text_truncated }.
navigate	Navigation	Open a URL in the active Playwright page.	{ ok, url, title }.
scroll	Navigation	Scroll the current page up or down by pixels.	{ ok, direction, pixels }.
wait_for	Navigation	Wait for a selector, visible text, or timeout.	{ ok, mode, selector/text/seconds }.
click	Interaction	Click the first element matching a CSS selector.	{ ok, selector, url }.
type_text	Interaction	Type text into the first input-like element matching a CSS selector.	{ ok, selector, chars }.
press_key	Interaction	Press a keyboard key such as Enter, Tab, or Escape.	{ ok, key, url }.
take_screenshot	Artifacts	Capture a PNG screenshot of the full page or one selector.	{ ok, file_name, path, url, title, size_bytes, full_page, selector }.
list_artifacts	Artifacts	List top-level runtime artifact files in /data.	{ ok, count, files }.
get_file	Artifacts	Read an artifact or download by safe file name and return base64 content.	{ ok, file_name, category, mime_type, size_bytes, content_base64 }.
get_downloaded_file	Artifacts	Read a downloaded file by safe file name when it exists in the data directory.	{ ok, file_name, category, mime_type, size_bytes, content_base64 }.

Most tools return a structured dictionary with ok: true on success. Expected tool-boundary failures return ok: false, error, and error_type.

browse_and_act is intentionally kept as a string-output tool for older MCP clients. Prefer browse_and_act_structured when your client can consume structured results.