Eight tools. JSON in, JSON out.

Fetch any public URL and get back clean Markdown, the page title and canonical metadata — no HTML parsing in your prompt.

Pass a URL and a JSON Schema; get back typed JSON that validates against it, with a confidence score per field.

Run a web search and return ranked results with titles, snippets and URLs your agent can pass straight to fetch_page.

Crawl a site from a start URL with depth and page limits. Returns a job id; results stream to get_job as pages finish.

Check the status of a crawl job and page through its results. Safe to poll; polling is never billed.

Render a page in a headless browser and return a short-lived PNG URL — useful when layout matters more than text.

Store any result in your workspace under a key so later runs can diff against it instead of fetching again.

Return billable events, spend and remaining credits for the current period, so an agent can check its own budget.