MCP tools

Twenty-five tools: one per API capability, plus the extra verbs crawls, batches and monitors need. Your agent reads these descriptions and picks the right one, so you rarely need to name a tool yourself. All parameters are snake_case. Most tools require only url. The exceptions: name for prefetch_resolve, descriptor for prefetch_transaction, query for prefetch_search, task for prefetch_answer, urls for the batch tools, domains for prefetch_logo_urls, id for the crawl and monitor tools that act on one, and name plus schedule to create a monitor. See MCP server for setup.

Finding a domain

`prefetch_resolve`

Turns a company, brand or product name into the domain of its official website: “Stripe” → stripe.com. Every other tool takes a URL, so this is where a task that starts from a name begins: resolve the name, then call prefetch_enrich, prefetch_brand or prefetch_company with the domain. Each result has a confidence from 0 to 1. Only the first can reach 0.8, and at or above it that result is the answer. Over MCP, names the index is unsure of are looked up live by default. That takes a few seconds and costs 3 credits, versus 1 from the index.

Parameter	Type	Default	Description
`name`	string	none	Required. The name, like `"Delta Air Lines"` or `"Figma"`. 1–100 characters.
`limit`	number	`5`	How many candidates to return. 1–10.
`live`	boolean	`true`	Look the name up live when the index is unsure. `false` for an instant, index-only lookup (free when nothing matches).

`prefetch_transaction`

Turns a card transaction descriptor (a bank statement line like SQ *BLUE BOTTLE COFFE OAKLAND CA) into the merchant behind it: its name, website, processor and location, and a logo_url ready to show. Lines that are payments to a person, transfers, fees or ATM withdrawals say so in kind. See GET /transaction.

Parameter	Type	Default	Description
`descriptor`	string	none	Required. The statement line. Up to 200 characters.
`country`	string	none	Two-letter code of the country the card was used in.
`mcc`	string	none	The four-digit merchant category code, if known.
`include_brand`	boolean	`false`	Also return the merchant’s logos and colors, as `prefetch_brand` does. +10 credits.

3 credits per lookup, 1 for a line looked up in the last 7 days.

Web extraction

`prefetch_search`

Search the web (Google results) and get each result’s title, URL and snippet, optionally with every result page scraped into markdown in the same call. Use it to find pages. For a researched answer, use prefetch_answer. See GET /search.

Parameter	Type	Default	Description
`query`	string	none	Required. What to search for. Search operators work: `"exact phrase"`, `-word`, `filetype:pdf`.
`limit`	number	`10`	Results to return. 1–50, or 1–10 with `scrape_results`.
`type`	string	`web`	`web` or `news`.
`time_range`	string	none	`hour`, `day`, `week`, `month` or `year`.
`country`, `language`	string	none	Two-letter codes.
`include_domains`, `exclude_domains`	string[]	none	Up to 10 each.
`scrape_results`	boolean	`false`	Also fetch each result page. 1 credit per page, 4 with `summary` or `json`.
`formats`	string[]	`["markdown"]`	With `scrape_results`: the page formats.

2 credits per 10 results, plus 1 per page scraped (4 with summary or json).

`prefetch_scrape`

Fetch one URL and return its content. Handles JavaScript-rendered pages, strips boilerplate, and resolves every link and image to an absolute URL. Use it when you know which page you want. To find pages first, use prefetch_map.

Parameter	Type	Default	Description
`url`	string	none	Required. `https://` is assumed if the scheme is omitted.
`formats`	string[]	`["markdown"]`	Any of `markdown`, `summary`, `html`, `raw_html`, `links`, `images`, `json`, `screenshot`.
`only_main_content`	boolean	`true`	Strip navigation, sidebars, headers, and footers.
`include_links`	boolean	`true`	Keep hyperlinks in the markdown.
`include_images`	boolean	`true`	Keep images in the markdown.
`include_base64_images`	boolean	`false`	Keep inline `data:` image URIs. These are very large.
`include_selectors`	string[]	none	CSS selectors to keep, overriding main-content detection.
`exclude_selectors`	string[]	none	CSS selectors to drop, like `[".cookie-bar", "#promo"]`.
`json_prompt`	string	none	For `formats: ["json"]`: what to extract, in plain English.
`json_schema`	object	none	For `formats: ["json"]`: a JSON Schema the result must match.
`summary_max_words`	number	`120`	Word budget for `formats: ["summary"]`. 20–400.
`screenshot_full_page`	boolean	`false`	For `formats: ["screenshot"]`: capture the whole page rather than the first screen.
`wait_for`	number	none	Milliseconds to wait after load for late content (up to 30,000).
`wait_for_selector`	string	none	Wait up to 10 seconds for this CSS selector before reading. Best effort.
`mobile`	boolean	`false`	Render the page as a phone.
`max_age`	number	none	Accept a stored copy fetched at most this many milliseconds ago. Default: always fetch live.
`headers`	object	none	Request headers to send to the site, like `{"Accept-Language": "de-DE"}`. Requests with headers are never cached.
`check_prompt_injection`	boolean	`false`	Check the page for text aimed at AI models and put the verdict above the page. Adds the 3-credit AI surcharge.

include_selectors is the best way to cut a large page down to what matters. It beats truncation, because you choose what survives.

raw_html is much larger than markdown. Only ask for it when you need the source.

`prefetch_map`

Discover the URLs on a site, from its sitemaps and on-page links, without fetching page content. Fast and cheap. Mapping first and then scraping the two or three pages you want is usually better, and much cheaper, than crawling.

Parameter	Type	Default	Description
`url`	string	none	Required.
`search`	string	none	Rank and filter results by this term, like `"pricing"`.
`limit`	number	`1000`	Maximum URLs to return. 1–5000.
`sitemap`	string	`include`	`include` merges sitemap and link discovery, `only` trusts the sitemap alone, `skip` ignores it.
`include_subdomains`	boolean	`true`	Include URLs on subdomains.
`ignore_query_params`	boolean	`true`	Collapse URLs that differ only by query string.

`prefetch_crawl`

Crawl a site or a section of one, returning each page as markdown. Starts the crawl and waits for it. See how crawling works. Crawls are billed per page. Scope them with include_paths and keep limit as low as will do.

Parameter	Type	Default	Description
`url`	string	none	Required.
`limit`	number	`25`	Maximum pages to crawl. 1–500. Each page is billed.
`max_depth`	number	`2`	How many links deep to follow. 0–5.
`include_paths`	string[]	none	Regex patterns; a URL must match at least one, like `["^/docs/"]`.
`exclude_paths`	string[]	none	Regex patterns; a URL matching any is skipped.
`allow_subdomains`	boolean	`false`	Follow links onto subdomains.
`allow_external_links`	boolean	`false`	Follow links onto other sites. This can make a crawl very large.
`ignore_query_params`	boolean	`true`	Treat URLs differing only by query string as one page.
`use_sitemap`	boolean	`true`	Seed the crawl from the site’s sitemap as well as from links.
`concurrency`	number	`2`	Pages fetched in parallel. 1–5.
`delay_ms`	number	`0`	Politeness delay between pages.
`scrape_options`	object	none	How each page is rendered: the same options `prefetch_scrape` takes.
`wait_seconds`	number	`60`	How long to wait before handing back a crawl id to poll. 0–120.

Sitemap-seeded pages sit at depth 0, because they were handed over rather than followed. So max_depth: 0 with use_sitemap: true crawls exactly what the sitemap lists and follows nothing.

`prefetch_check_crawl_status`

Status and results for a crawl or batch. Returns one page of results at a time.

Parameter	Type	Default	Description
`id`	string	none	Required. The id from `prefetch_crawl`, `prefetch_batch_scrape` or `prefetch_batch`.
`skip`	number	`0`	Results to skip, for paging.
`limit`	number	`25`	Results per page. 1–100.

Calling this is free. Only pages that completed since your last call are billed.

`prefetch_cancel_crawl`

Stop a running crawl. Pages already fetched stay readable via prefetch_check_crawl_status until the crawl expires.

Parameter	Type	Default	Description
`id`	string	none	Required. The crawl id from `prefetch_crawl`.

`prefetch_batch_scrape`

Scrape many known URLs in one job and get each page as markdown. Starts the job and waits for it (wait_seconds, default 60), returning the pages if it finishes in time and an id for prefetch_check_crawl_status if not. See POST /batch/scrape.

Parameter	Type	Default	Description
`urls`	string[]	none	Required. 1–500 pages.
`scrape_options`	object	none	How each page is rendered: the options `prefetch_scrape` takes.
`concurrency`	number	`2`	Pages fetched in parallel. 1–5.
`ignore_invalid_urls`	boolean	`false`	Drop invalid URLs instead of failing the batch.

1 credit per page returned.

`prefetch_batch`

Run brand, enrich, company, classify, screenshot or styleguide on a list of URLs or bare domains as one job, and get every result back: the same result each single tool returns. Waits like prefetch_batch_scrape. See POST /batch/{endpoint}.

Parameter	Type	Default	Description
`endpoint`	string	none	Required. Which tool to run on every URL.
`urls`	string[]	none	Required. 1–500 URLs or domains, like `["stripe.com", "linear.app"]`.
`options`	object	none	That tool’s own options, applied to every URL: `{"design_tokens": true}` for brand, `{"device": "iphone_15"}` for screenshot.
`concurrency`	number	`2`	URLs run in parallel. 1–5.
`ignore_invalid_urls`	boolean	`false`	Drop invalid URLs instead of failing the batch.

Each URL costs what one call of that tool costs; failed URLs are free.

Brand and company data

`prefetch_enrich`

Everything known about the company behind a URL in one call: brand assets, company details, and IAB classification.

Prefer this over calling prefetch_brand, prefetch_company, and prefetch_classify separately. It is one page fetch instead of three, and costs less than the three combined.

Parameter	Type	Default	Description
`url`	string	none	Required.
`html`	boolean	`false`	Also extract colors from inline HTML style attributes.
`css`	boolean	`false`	Also extract colors from stylesheets, grouped by CSS property.
`javascript`	boolean	`false`	Also extract colors from inline JavaScript.
`screenshot_colors`	number	none	Sample this many dominant colors from a rendered screenshot. 2–12.
`screenshot_width`	number	`1440`	Screenshot viewport width.
`screenshot_height`	number	`900`	Screenshot viewport height.
`screenshot_full_page`	boolean	`false`	Capture the whole scrollable page.
`filter_achromatic`	boolean	`false`	Drop near-white, near-black, and grey results.
`design_tokens`	boolean	`false`	Also return `brand.design_tokens`: colors by role, heading and body type, and button and input styles.

screenshot_colors catches colors that only exist inside images, at the cost of a slower call. design_tokens is what to ask for when building UI that should match the site.

`prefetch_brand`

Brand identity for a URL: palette with color names and hex values, font families, logos, favicon, and Open Graph images. Takes the same parameters as prefetch_enrich. Use prefetch_enrich instead if you also want company details or industry.

`prefetch_company`

Company name, contact emails, postal addresses, and social profiles for the organization behind a URL.

Parameter	Type	Default	Description
`url`	string	none	Required.

`prefetch_classify`

IAB content category and subcategory, with a description and, for public companies, stock tickers and ISINs. The cheapest way to answer “what kind of business is this”.

Parameter	Type	Default	Description
`url`	string	none	Required.

`prefetch_screenshot`

Render a page and return a hosted image URL.

Parameter	Type	Default	Description
`url`	string	none	Required.
`width`	number	`1440`	Viewport width. 320–2560. Ignored with `device`.
`height`	number	`900`	Viewport height. 240–5000. Ignored with `device`.
`format`	string	`jpeg`	`jpeg`, `png`, `webp`, or `pdf`.
`full_page`	boolean	`false`	Capture the whole scrollable page.
`device`	string	none	Render as a phone, tablet or retina desktop: `iphone_15`, `pixel_7`, `ipad`, `desktop_retina`… (see `/screenshot`).
`dark_mode`	boolean	`false`	Ask the page for its dark theme.
`selector`	string	none	Capture only the first element matching this CSS selector.
`wait_for`	number	`0`	Milliseconds to wait after load. Up to 30,000.
`wait_for_selector`	string	none	Wait (up to 10 s) for this CSS selector before capturing.
`block_ads`	boolean	`false`	Block ad and tracker requests while the page loads.
`quality`	number	`85`	JPEG and WebP quality. 1–100.

This returns a link, not image bytes. Pass the URL on to the user, or to a tool that can fetch images.

`prefetch_styleguide`

A site’s design system, read from the rendered page: colors by role, the type scale, spacing, radii, shadows, button, input, card and link styles, breakpoints and the site’s own CSS variables, plus the same as CSS custom properties (exports.css) and a Tailwind theme (exports.tailwind). Use it to build UI that matches a site. See GET /styleguide.

Parameter	Type	Default	Description
`url`	string	none	Required.
`dark_mode`	boolean	`false`	Read the site’s dark theme instead of its default one.

10 credits.

`prefetch_logo_urls`

Ready-to-embed logo image URLs for up to 50 domains, from the logo CDN. Put them straight in <img> tags or markdown. A domain without a known logo shows a letter mark.

Parameter	Type	Default	Description
`domains`	string[]	none	Required. 1–50 domains or URLs.
`size`	number	`128`	Pixels, square. 16–512.
`format`	string	`png`	`png`, `webp` or `jpeg`.

Free: logos use your plan’s monthly logo quota, not credits.

Research

`prefetch_answer`

Research a question on the live web and get a sourced answer as JSON in the shape you describe. It searches the web, reads pages, and looks up company records as the task needs, then returns json_content in the shape of json_format plus the sources it read. Facts it could not establish come back null. Use it when the answer needs searching or several sources. When you already know the page, prefetch_scrape with formats: ["json"] is cheaper; for a company’s profile from its URL, use prefetch_enrich.

Parameter	Type	Default	Description
`task`	string	none	Required. The research question. Name a domain or URL to focus it: named pages are read first. Up to 2,000 characters.
`mode`	string	`fast`	`fast`: up to 30 s, 10 credits. `ultra`: deeper research across several sources, up to 50 s, 100 credits.
`json_format`	object	`{"result": ""}`	An example of the answer object, not a JSON Schema, like `{"ceo": "", "founded_year": 0}`. Keys and value types are kept; unknowns come back `null`.
`timeout_opts`	object	none	`{ milliseconds, behavior }`. Shortens the mode’s deadline (1,000–300,000 ms). With `behavior: "return-partial"`, what was found so far is returned with `partial: true`.
`tags`	string[]	none	Labels stored on the request log, for filtering usage. Up to 20.

Over MCP the default mode is fast; over REST it is ultra. Failures and timeouts are never charged.

Monitoring

A monitor re-checks pages on a schedule and records what changed. See Monitors for how changes are detected and judged. The tools take a flatter shape than the REST body: urls for specific pages and crawl_url for a site, instead of a list of targets. An agent rarely has a webhook to receive changes at, so it reads them back, or passes email to have each check that finds changes emailed to a person: prefetch_run_monitor runs a check and waits for it, and prefetch_get_monitor_check shows the latest check by default. Both list only the pages that changed, appeared, disappeared or failed, the judge’s verdict first, then any field changes, then the diff, and count the unchanged ones.

`prefetch_create_monitor`

Start watching pages. Pass urls, crawl_url, or both. A monitor bills on every check for as long as it’s active: 1 credit per page fetched, +3 when a JSON extraction runs, and +1 per change judged. So prefer few pages and a modest schedule.

Parameter	Type	Default	Description
`name`	string	none	Required. A label, like `"Acme pricing"`.
`schedule`	string	none	Required. A phrase: `"every 30 minutes"`, `"hourly"`, `"daily at 9am"`, `"every weekday at 17:00"`, `"weekly"`, or five-field cron. As often as the plan allows: hourly on Free, every 5 minutes on Starter and Pro, 2 on Growth, 1 on Scale.
`timezone`	string	`UTC`	IANA timezone for the schedule, like `"America/New_York"`.
`goal`	string	none	What you want to be told about, in one sentence. Each change is judged against it, so noise is filtered out.
`urls`	string[]	none	Pages to watch. 1–100.
`crawl_url`	string	none	A site or section to re-crawl on every check, so new and removed pages are reported too.
`crawl_limit`	number	`25`	Most pages each crawl visits. 1–500. Every page is billed on every check.
`include_paths`	string[]	none	For `crawl_url`: regex patterns a page must match, like `["^/blog/"]`.
`exclude_paths`	string[]	none	For `crawl_url`: regex patterns to skip.
`track_fields`	string	none	Compare these facts instead of the page text, like `"each plan's monthly price"`. A page then counts as changed only when one of them changes.
`include_selectors`	string[]	none	CSS selectors to compare instead of the whole main content.
`exclude_selectors`	string[]	none	CSS selectors to ignore: timestamps, rotating banners.
`webhook_url`	string	none	An HTTPS endpoint to POST changes to.
`email`	string[]	none	Up to 5 addresses to email when a check finds changes. Your account’s own address works at once; any other address confirms from an email first.
`run_now`	boolean	`false`	Take the first check immediately. It records a baseline (every page is new), so changes show from the second check on.

`prefetch_run_monitor`

Run a check now, wait for it, and return what changed since the previous check. Billed like any check.

Parameter	Type	Default	Description
`id`	string	none	Required. The monitor id.
`wait_seconds`	number	`60`	How long to wait before handing back the check id to read later. 0–120.

`prefetch_get_monitor_check`

What a check found. Leave check_id out for the most recent one. Free.

Parameter	Type	Default	Description
`monitor_id`	string	none	Required.
`check_id`	string	latest	A check id.
`status`	string	none	Only pages with this status: `same`, `new`, `changed`, `removed` or `error`. Without it, unchanged pages are counted but not listed.
`skip`	number	`0`	Pages to skip, for paging.
`limit`	number	`25`	Pages per response. 1–100.

`prefetch_list_monitors`

The key’s monitors with each one’s schedule and last check. Free.

Parameter	Type	Default	Description
`limit`	number	`25`	Monitors per page. 1–100.
`offset`	number	`0`	Monitors to skip.

`prefetch_get_monitor`

One monitor’s settings, next run, estimated cost, and its five most recent checks. Free.

Parameter	Type	Default	Description
`id`	string	none	Required.

`prefetch_update_monitor`

Change a monitor. Fields you leave out stay as they are. Free.

Parameter	Type	Default	Description
`id`	string	none	Required.
`name`	string	none	A new label.
`schedule`, `timezone`	string	none	A new schedule. Pass `timezone` together with `schedule`.
`goal`	string \| null	none	A new goal, or `null` to stop judging and report every change.
`status`	string	none	`paused` stops the schedule and its charges; `active` resumes it.
`urls`, `crawl_url`, …		none	Replace what the monitor watches. Takes the same fields as `prefetch_create_monitor`.
`webhook_url`	string \| null	none	A new webhook URL, or `null` to stop sending webhooks.
`email`	string[] \| null	none	The full new list of addresses to email, replacing the old one. `null` or `[]` stops emailing.

`prefetch_delete_monitor`

Delete a monitor and its history, stopping all future checks. To stop it for a while, pause it with prefetch_update_monitor instead. Free.

Parameter	Type	Default	Description
`id`	string	none	Required.

Choosing a tool

You want	Use	Credits
Pages about something	`prefetch_search`	2 per 10 results, +1 per page scraped
One page’s content	`prefetch_scrape`	1
To know which pages exist	`prefetch_map`	1
A few specific pages	`prefetch_map`, then `prefetch_scrape` on each	1 + 1 each
A whole section of a site	`prefetch_crawl` with `include_paths`	1 per page
Many known pages	`prefetch_batch_scrape`	1 per page
The merchant behind a card transaction	`prefetch_transaction`	3
Companies’ logos to show	`prefetch_logo_urls`	0
To build UI that matches a site	`prefetch_styleguide`	10
The same profile for many companies	`prefetch_batch` with `endpoint: "enrich"`	12 per company
Brand, company, and industry	`prefetch_enrich`	12
Just the brand colors	`prefetch_brand`	10
Just the industry	`prefetch_classify`	5
A picture of the page	`prefetch_screenshot`	1
An answer that needs searching	`prefetch_answer`	10, or 100 in `ultra` mode
To know when a page changes	`prefetch_create_monitor`	1 per page, every check
What changed, right now	`prefetch_run_monitor`	1 per page, plus 1 per change judged

Get started

Agents and MCP

Company data

Web data

Jobs and monitoring

Platform

Going to production

Finding a domain

`prefetch_resolve`

`prefetch_transaction`

Web extraction

`prefetch_search`

`prefetch_scrape`

`prefetch_map`

`prefetch_crawl`

`prefetch_check_crawl_status`

`prefetch_cancel_crawl`

`prefetch_batch_scrape`

`prefetch_batch`

Brand and company data

`prefetch_enrich`

`prefetch_brand`

`prefetch_company`

`prefetch_classify`

`prefetch_screenshot`

`prefetch_styleguide`

`prefetch_logo_urls`

Research

`prefetch_answer`

Monitoring

`prefetch_create_monitor`

`prefetch_run_monitor`

`prefetch_get_monitor_check`

`prefetch_list_monitors`

`prefetch_get_monitor`

`prefetch_update_monitor`

`prefetch_delete_monitor`

Choosing a tool

​Finding a domain

​prefetch_resolve

​prefetch_transaction

​Web extraction

​prefetch_search

​prefetch_scrape

​prefetch_map

​prefetch_crawl

​prefetch_check_crawl_status

​prefetch_cancel_crawl

​prefetch_batch_scrape

​prefetch_batch

​Brand and company data

​prefetch_enrich

​prefetch_brand

​prefetch_company

​prefetch_classify

​prefetch_screenshot

​prefetch_styleguide

​prefetch_logo_urls

​Research

​prefetch_answer

​Monitoring

​prefetch_create_monitor

​prefetch_run_monitor

​prefetch_get_monitor_check

​prefetch_list_monitors

​prefetch_get_monitor

​prefetch_update_monitor

​prefetch_delete_monitor

​Choosing a tool

Finding a domain

`prefetch_resolve`

`prefetch_transaction`

Web extraction

`prefetch_search`

`prefetch_scrape`

`prefetch_map`

`prefetch_crawl`

`prefetch_check_crawl_status`

`prefetch_cancel_crawl`

`prefetch_batch_scrape`

`prefetch_batch`

Brand and company data

`prefetch_enrich`

`prefetch_brand`

`prefetch_company`

`prefetch_classify`

`prefetch_screenshot`

`prefetch_styleguide`

`prefetch_logo_urls`

Research

`prefetch_answer`

Monitoring

`prefetch_create_monitor`

`prefetch_run_monitor`

`prefetch_get_monitor_check`

`prefetch_list_monitors`

`prefetch_get_monitor`

`prefetch_update_monitor`

`prefetch_delete_monitor`

Choosing a tool