Broken Link Checker — same-host crawl and failing links

Give it a start URL or a sitemap; it crawls pages of the same site (up to your page limit, obeying robots.txt, at most two requests a second), checks every link and asset once, and returns each broken or redirected link with its status, redirect chain, anchor text and the pages it was found on, plus a summary.

Development validation produced actual successful public-input records using the production Dockerfile and SDK entrypoint on GitHub-hosted Linux. The smoke input, all three tasks and every defined benchmark passed their exact expectations. The independent review completed on 2026-10-19; its findings are tracked in the committed fix report. Platform costs, settlement, publication and daily tests remain unverified.

Call it from an agent

Through Apify's MCP server (https://mcp.apify.com) use call-actor:

{
  "actor": "hermeskd/broken-links",
  "input": {
    "startUrl": "https://docs.apify.com/platform/actors/publishing/monetize",
    "maxPages": 1,
    "checkExternal": false
  },
  "callOptions": {
    "maxTotalChargeUsd": 0.05,
    "timeout": 3600
  },
  "waitSecs": 30
}

Then call get-dataset-items with the returned defaultDatasetId. Dedicated tool URL: https://mcp.apify.com?tools=hermeskd/broken-links

Input

Field	Type	Meaning
startUrl	string	Exactly one startUrl or sitemapUrl is required.
sitemapUrl	string	Alternative sitemap entry point; same-host pages only.
maxPages	integer	Maximum successfully parsed same-host HTML pages.
checkExternal	boolean	Check external links but never crawl external pages.
reportRedirects	boolean	Return successful redirected link findings, in addition to failures.

Output

Field	Meaning
type	Link finding or final crawl summary.
url	Checked URL, or crawl entry point on summary/error records.
status	Final HTTP status; null when no response.
ok	False for broken/error records. These are never charged.
redirectChain	Redirect destination URLs in observed order.
anchorText	First anchor text observed, capped at 500 characters.
foundOn	At most ten referring page URLs.
summary	All attempted unique checks and successful parsed page counts; linksChecked is not billable count.
error	Failure, always uncharged.

Actual production-container record from 2026-10-18T20:25:14.282887+00:00 (code f063187693578dd7551fa2bec552281e2f0f3e33, smoke input, SDK local storage). The saved record is unchanged; local event logs do not prove Apify settlement.

{
  "type": "summary",
  "url": "https://docs.apify.com/platform/actors/publishing/monetize",
  "ok": true,
  "summary": {
    "pagesCrawled": 1,
    "linksChecked": 53,
    "broken": 0,
    "redirected": 1
  }
}

The saved example above records the October 18 source and its exact code pin. The approved live page changed by October 19: both reviewed and repaired extractors find 52 same-host targets, confirmed by all five new production-container cases. The manifest retains exact checks for that observed inventory; the committed fix report records the source hashes and final validation. This does not change prices or make failed checks billable.

Events and prices

Event	Title	Charged when	FREE / BRONZE	SILVER	GOLD / PLATINUM / DIAMOND
link-checked	Link checked	One successful unique link or asset check, backed by a readable successful summary.	$0.0005	$0.00045	$0.0004
page-crawled	Page crawled	One successfully fetched and parsed site page, backed by a readable successful summary.	$0.001	$0.0009	$0.0008

Platform usage is not passed through. Apify's automatic actor-start event remains $0.00005 per started GB (one GB or less counts once). The synthetic dataset-item event is absent. Charge limits stop further paying work; unprocessed inputs receive uncharged errors. Completed readable pending events are settled on abort.

Worked example: two successful unique checks and one 404 on one parsed page cost $0.002 at list ($0.001 for two checks plus $0.001 for the page), plus actor start. linksChecked is three; link-checked events are two. Duplicate checks, broken/error findings and failed/unreadable summaries add no events.

Limits and run time

One startUrl or sitemapUrl; maxPages 1–1,000. Same-host HTML crawling only; external links may be checked but are not crawled. At most 100,000 unique checks, 10 MiB per fetched page/GET fallback, ten foundOn pages per finding. HEAD falls back to GET for 405 or 501; an ordinary bodyless successful HEAD is sufficient. Fragments and duplicate targets share one check. All HTTP requests are at most two per host per second. linksChecked counts attempts and is not the billable count. Broken findings, policy refusals, transport failures and errors are never charged. A failed/timeout or unreadable summary backs no charges, including previously completed checks.

Default run timeout is 3600 seconds. Typical cold platform time, costs and the smallest passing memory have not been measured for this new implementation. Memory probes are evidence to collect later; configured memories are specification defaults, not measured claims.

Errors

Code	Meaning
login_required	The page requires authentication or is a sign-in wall; no authentication was attempted.
robots_disallowed	robots.txt forbids fetching the URL; no request was made to it.
blocked_host	The host is on the deny-list or resolves to a private or local address; no request was made.
download_failed	The request or connection failed within its bounded deadline.
http_error	The source answered an unsuccessful HTTP status.
unsupported_content_type	An HTML audit/crawl received a non-HTML representation.
too_large	A documented response, expansion or traversal cap was exceeded.
source_schema_changed	A source or tool response no longer has the verified protocol shape.
timeout	The input did not finish within the run's time budget.
invalid_input	The run input passed the input schema but cannot be used; the message names the field.
internal_error	Unexpected failure; the message carries the error class.
charge_limit_reached	The run reached its maximum total charge; this input was not processed.

No failure record is charged. A run fails only when all input outcomes are Actor-side failures; caller errors and a mixture of caller/Actor failures do not fail the run.

Sources and terms

Only process content you have the right to process. URL fetches obey HermesBot robots-group selection, the shared deny-list, public-address checks on connections and redirects, honest identity, deadlines and rate limits. No caller credentials, cookies, CAPTCHA solving, residential proxies or evasion. Only same-host HTML pages are crawled. External links may be checked when requested, and are never crawled. Sign-in walls are refused without authentication. Broken and error findings are free; successful checks and parsed pages are charged only after a successful readable summary. A failed or unreadable summary backs no charges. Initial source HEAD observations were expectations. The 2026-10-18 GitHub-hosted production-container crawl produced the real records shown above, through the guarded SDK path. Responsive-image srcset candidates, picture/source srcset and video poster assets now use the same unique checks; every failed check remains free.

Not affiliated

This Actor is not affiliated with Google, Apify, IANA, Verisign or other named protocol/source providers. Third-party names describe compatibility and source material.

Related Hermes Actors

website-screenshot, lighthouse-audit, seo-page-audit, domain-availability, dns-and-email-auth, broken-links and sitemap-urls.

Changelog

2026-10-17 (historical): initial implementation and fixture contracts; packaging and public production validation were incomplete on that date.

2026-10-18: all six production images, public smoke inputs, three tasks per Actor and defined benchmarks passed on GitHub-hosted Linux; exact code, records and image provenance are retained with the examples. These local SDK runs do not prove platform costs, settlement, publication, minimum memory, daily tests or payouts.

2026-10-19: independent-review repairs cover DNS privacy, malformed publisher URLs, empty sitemap failures, responsive assets and schema prefill, as applicable to this Actor. See the committed unit 4 fix report for measured verification and remaining platform NO-COVERAGE.