Recurring Web Check for AI Agents
Pricing
from $3.00 / 1,000 results
Checks many URLs in one run. Per URL returns clean Markdown, CSS/JSON-extracted fields and a diff vs. the previous run under the same watchId, so only what changed is flagged. For agents doing recurring checks: prices, docs, changelogs, status pages, job boards, JSON APIs. No login, browser or LLM.
Absolute URLs to check in this run. Put ALL URLs of one recurring job here — one run handles many URLs, you do not need one run per URL. HTML pages, JSON/REST endpoints, RSS/Atom feeds and plain-text files all work.
monitor (default): remembers each URL per watchId and reports changed=true with a compact diff when the page or any field changed since the last run. fetch: stateless, returns content + fields. extract: stateless, returns only the fields (tiny output).
Stable name for a recurring job, e.g. competitor-pricing or docs-acme-api. Snapshots are stored under this ID, so every later run with the same watchId diffs against the previous one. Leave empty and an ID is derived from the URL set + selector + fields, so identical repeat calls still get change detection.
Optional map of field name → extraction spec. Specs: 'h1' (text of first match), 'a.buy@href' (attribute), '.price | /([\d.,]+)/' (regex on the text, first capture group), 'all:.item .name' (array of all matches), 'json:data.items[0].price' (path into a JSON response). Changed field values are reported in fieldChanges with before/after. No LLM involved — deterministic and free of hallucination.
Optional CSS selector that limits the content to one part of the page, e.g. 'main', '#pricing', 'article'. Smaller scope = fewer tokens and fewer false 'changed' flags. If it matches nothing, the whole body is used and scopeMatched=false is returned.
Elements to remove before comparing, e.g. ads, comments, timestamps, counters, 'related posts'. Use this to stop noisy page parts from triggering changed=true.
Regular expressions removed from the text before storing and comparing, e.g. clock times, '5 minutes ago', session IDs. Plain pattern or /pattern/flags.
markdown keeps headings/lists/tables so an LLM can still see structure; text is flatter and a bit smaller.
Token budget: page content returned per URL is cut at this many characters (truncated=true tells you when). Set 0 to omit content entirely and only receive fields + diff. Change detection always uses the FULL content, not the truncated one.
compact returns a short +/- line diff of what changed; none returns only the change flags and field changes.
Upper bound for the diff text per URL.
Minimum share of changed characters (0–1) before content counts as changed. 0 = any difference. 0.02 ignores tiny edits like a rotated testimonial. Field changes always count regardless of this threshold.
Remove nav, header, footer, aside, forms and cookie banners from the content (ignored when a scope selector is set).
By default links are reduced to their text (URLs with tracking parameters change on every load and inflate output). Enable to keep text.
Optional. When at least one URL changed, a JSON payload {event, watchId, runId, changed:[...]} is POSTed here — handy for scheduled runs that should wake up an agent only when something happened.
Optional. Leave empty for direct requests. Enable Apify Proxy for sites that block datacenter IPs.
Parallel requests. Lower it for sites with strict rate limits.
Per-URL timeout. Keep runs fast: an MCP call-actor waits at most 45 s for the whole run.
