git-annex can access a remote using any web server, as shown in the tip setup a public repository on a web site.

That's limited to basic read-only repository access though. Git has smart HTTP that can be used to allow pushes over http. And git-annex has the git-annex p2phttp command.

As well as allowing write access to authorized users over http, git-annex p2phttp also allows accessing clusters, and other proxied remotes over http.

You will still need to run a web server to serve the git repository. git-annex p2phttp only serves git-annex's own API, and it does it on a different port (9417 by default).

You will need to arrange to run git-annex p2phttp in your repository as a daemon or service. Note that it should not be run as root, but as whatever user owns the repository. It has several options you can use to configure it, including controlling who can access the repository.

So there are two web servers, and thus two different urls. A remote will have remote.name.url set to the http url that git will use, and also have remote.name.annexUrl set to the url that git-annex will use to talk to git-annex p2phttp. That url looks like this:

annex+http://example.com/git-annex/

The "annex+http" (or "annex+https") indicates that it's a git-annex API url, which defaults to being on port 9417 unless a different port is set.

It would be annoying if every user who cloned your repository had to set remote.name.annexUrl manually. So there's a way to automate it. In the git config file of the repository, set annex.url to the "annex+http" (or "annex+https") url. The first time it uses a http remote, git-annex downloads its git config file, and sets remote.name.annexUrl to the value of the remote's annex.url.

Support for this first appeared in git-annex version 10.20240731. Users of older git-annex won't be able to use the smart http server, but can still clone the repository from the dumb http server.