Read and extract

Markdown from pages and URLs, tables, links, structured records, and WebMCP tools.

A page you are on

browser_read takes what:

what	Returns
text (default)	Visible text of the page or of one ref
markdown	The main content converted to markdown: headings, lists, links, code blocks and tables. Navigation, footers and ads are dropped
links	Every link with its text and absolute URL
tables	Each table as an array of rows
html	Raw outer HTML

A URL you are not on

browser_fetch {url, what} opens a background tab in your window, waits for load, reads the main content as markdown (or text or HTML) and closes the tab. It is the fast path for “read this documentation page and summarise it”. Pass keepTab: true to leave the tab open for follow-up actions, or context to fetch inside a developer context instead.

Because it is a real browser tab, pages that need JavaScript, logins from your session, or your cookies render correctly, unlike a plain HTTP fetch.

Structured extraction

browser_extract turns repeated elements into records:

{
  "items": ".product-card",
  "fields": {
    "name": "h3",
    "price": ".price",
    "link": { "selector": "a", "attr": "href" },
    "image": { "selector": "img", "attr": "src" }
  },
  "limit": 100
}

Each field is a CSS selector evaluated inside the item, returning text, or {selector, attr} for an attribute. The result is a JSON array. Combine with browser_scroll or clicking “Load more” for paginated lists.

Tools the page exposes

Pages that implement WebMCP register typed tools on navigator.modelContext. browser_webmcp {action: "list"} reports them with their schemas, and {action: "invoke", name, args} calls one. When a page offers a searchProducts tool that beats scraping the results grid.

Saving what you read

• browser_pdf prints the page to a PDF file (A4, Letter or Legal, landscape, background graphics, scale) and returns the path, ~/.browspark/artifacts/… by default.
• browser_screenshot {fullPage: true} for an image.