Keep the page context
Extract the title, description, declared canonical URL and language. Supply the source URL to resolve relative links. Declared metadata can be incorrect; keep your original source URL too.
html.metadata · $0.0003 per call ↗Research workflows
Turn supplied HTML into structured research inputs
Navigation, repeated links and formatting make raw HTML awkward to use in a research workflow. Keep extraction separate from interpretation and preserve the source context.
$0.0011 for one successful call to each of the 3 tools below.
Extract readable content
Send the same supplied HTML to obtain article text and the extraction method. This parses HTML; it does not execute JavaScript or render content that is absent from the document.
html.main-text · $0.0005 per call ↗Keep bounded snippets and offsets
Split extracted text into overlapping chunks. Offsets use UTF-16 units, and surrogate pairs remain intact. This is character-based chunking, not a model-specific token count.
text.chunk · $0.0003 per call ↗What you get
Readable, bounded chunks alongside source metadata, ready for your own retrieval or summarization workflow. The agent supplies the HTML it is authorized to access.
These are separate API calls that your agent orchestrates, not an automatic bundled workflow. Review each tool's example and contract; input fields must be mapped explicitly between steps.
A small first step
Pick a useful tool.
Give it a real task.
Review its example and price, then create an API key and save your recovery key. Add $5 in service credits through Stripe and call the API from your agent.