Tools reference
Six tools, each a thin wrapper over its /v1 endpoint — same engine, same cache, same credits. Tools answer in text for the model to read, not JSON.
read_url
A page as clean Markdown — the default when the agent just needs to read.
/v1/read1 creditA public http(s) URL.
Skip the one-hour cache and fetch again. Default false.
Two text blocks: the Markdown, then the source's metadata — title, sizes, reduction, time taken, and whether it came from the cache.
{
"name": "read_url",
"arguments": {
"url": "https://stripe.com/pricing"
}
}read_and_score
The Markdown and how far to trust it, in one call.
/v1/read-and-score1 creditA public http(s) URL.
Skip the one-hour cache and fetch again. Default false.
The Markdown, then a quality block: score and grade, hallucination risk, reduction and cache state.
PDFs and plain text come back unscored.
{
"name": "read_and_score",
"arguments": {
"url": "https://stripe.com/pricing"
}
}score_url
Just the score, and why — for deciding whether a source is worth reading.
/v1/scorefreeA public http(s) URL.
One text block: score and grade, hallucination risk, what the page does well, its penalties, and recommendations.
Never cached — every call scores the page as it is now.
{
"name": "score_url",
"arguments": {
"url": "https://stripe.com/pricing"
}
}extract_data
The structured data a page already declares — facts you can rely on, with no model guessing.
/v1/extract1 creditA public http(s) URL.
Skip the one-hour cache and fetch again. Default false.
The page's JSON-LD, OpenGraph and meta tags, with counts of each.
{
"name": "extract_data",
"arguments": {
"url": "https://stripe.com/pricing"
}
}map_site
A site's URLs, without reading them — to plan what to read next.
/v1/map1 creditA public http(s) URL.
How many URLs, 1–1,000. Default 100.
The URLs, and where they came from: the sitemap, or the links on the start page. Same host only.
{
"name": "map_site",
"arguments": {
"url": "https://docs.stripe.com",
"limit": 200
}
}batch
Many pages in one call — a list, or a whole site.
/v1/batch1 per URL credits1–50 URLs. Wins if `site` is also given.
A site to discover and read instead of a list.
read, read-and-score or extract. Default read-and-score.
With `site`: how many pages, 1–50. Default 25.
One text list with a section per URL. Each page's Markdown is cut at 12,000 characters.
A credit per URL; URLs that fail are refunded.
{
"name": "batch",
"arguments": {
"urls": [
"https://stripe.com/pricing",
"https://stripe.com/about"
],
"mode": "read"
}
}Backed by the Read API
[02]Each tool calls the endpoint on its card, so the Read API reference has the full response each one is built from, and Error codes covers what can go wrong.