Build Status

Very fast link-checking.

linkcheck versus the popular blc tool


A good utility is custom-made for a job. There are many link checkers out there, but none of them seems to be striving for the following set of goals.

Crawls fast

  • You want to run the link-checker at least before every deploy (on CI or manually). When it takes ages, you're less likely to do so.

  • linkcheck is currently several times faster than blc and all other link checkers that go to at least comparable depth. It is 40 times faster than the only tool that goes to the same depth (linkchecker).

Finds all relevant problems

  • No link-checker can guarantee correct results: the web is too flaky for that. But at least the tool should correctly parse the HTML (not just try to guess what's a URL and what isn't) and the CSS (for url(...) links).

    • PENDING: srcset support
  • linkcheck finds more than linklint and blc. It finds the same amount or more problems than the best alternative, linkchecker.

Leaves out irrelevant problems

  • linkcheck doesn't attempt to render JavaScript. It would make it at least an order of magnitude slower and way more complex. (For example, what links and buttons should the tool attempt to click, and how many times? Should we only click visible links? How exactly do we detect broken links?) Validating SPAs is a very different problem than checking static links, and should be approached by dedicated tools.

  • linkcheck only supports http: and https:. It won't try to check FTP or telnet or nntp links.

  • linkcheck doesn't validate file system directories. Servers often behave very differently than file systems, so validating links on the file system often leads to both false positives and false negatives. Links should be checked in their natural habitat, and as close to the production environment as possible. You can (and should) run linkcheck on your localhost server, of course.

Good UX

  • Yes, a command line utility can have good or bad UX. It has mostly to do with giving sane defaults, not forcing users to learn new constructs, not making them type more than needed, and showing concise output.

  • The most frequent use cases should be only a few arguments.

    • For example, unleashing linkcheck on http://localhost:4001/ can be done via linkcheck :4001.
  • linkcheck doesn't throttle itself on localhost.

  • linkcheck follows POSIX CLI standards (no @input and similar constructs like in linklint).

Brief and meaningful output

  • When everything works, you don't want to see a huge list of links.

    • In this scenario, linkcheck just outputs 'Perfect' and some stats on a single line.
  • When things are broken, you want to see where exactly is the problem and you want to have it sorted in a sane way.

    • linkcheck lists broken links by their source URL first so that you can fix many links at once. It also sorts the URLs alphabetically, and shows both the exact location of the link (line:column) and the anchor text (or the tag if it wasn't an anchor).
  • For CI builds, you want non-zero exit code whenever there is a problem.

    • linkcheck returns status code 1 if there are warnings, and status code 2 if there are errors.

It goes without saying that linkcheck honors robots.txt and throttles itself when accessing websites.


Step 1. Install Dart

Full installation guides per platform:

For example, on a Mac, assuming you have homebrew, you just run:

$ brew tap dart-lang/dart
$ brew install dart

Step 2. Install linkcheck

Once Dart is installed, run:

$ pub global activate linkcheck

Pub installs executables into ~/.pub-cache/bin, which may not be on your path. You can fix that by adding the following to your shell's config file (.bashrc, .bash_profile, etc.):

export PATH="$PATH":"~/.pub-cache/bin"

Then either restart the terminal or run source ~/.bash_profile (assuming ~/.bash_profile is where you put the PATH export above).


If in doubt, run linkcheck -h. Here are some examples to get you started.


Running linkcheck without arguments will try to crawl http://localhost:8080/ (which is the most common local server URL).

  • linkcheck to crawl the site and ignore external links
  • linkcheck -e to try external links

If you run your local server on http://localhost:4000/, for example, you can do:

  • linkcheck :4000 to crawl the site and ignore external links
  • linkcheck :4000 -e to try external links

linkcheck will not throttle itself when accessing localhost. It will go as fast as possible.

Deployed sites

  • linkcheck to crawl and ignore external links
  • linkcheck to start directly on https
  • linkcheck to crawl both sites and check links between the two (but ignore external links outside those two sites)

Many entry points

Assuming you have a text file mysites.txt like this:

You can run linkcheck -i mysites.txt and it will crawl all of them and also check links between them. This is useful for:

  1. Link-checking projects spanning many domains (or subdomains).
  2. Checking all your public websites / blogs / etc.

There's another use for this, and that is when you have a list of inbound links, like this:

You probably want to make sure you never break your inbound links. For example, if a page changes URL, the previous URL should still work (redirecting to the new page when appropriate).

Where do you get a list of inbound links? Try your site's sitemap.xml as a starting point, and — additionally — try something like the Google Webmaster Tools’ crawl error page.