To my surprise, I discovered that ssh-copy-id fails if you do not have the private part of the key, making it not usable for deploying someone else key.
I an looking for a reliable way to deploy these keys, so it must not create duplicate entries if you run it twice.
Try this (newlines optional):
cat ~/.ssh/id_rsa.pub |
ssh <REMOTE> "cat > tmp.pubkey ;
mkdir -p .ssh ;
touch .ssh/authorized_keys ;
sed -i.bak -e '/$(awk '{print $NF}' ~/.ssh/id_rsa.pub)/d' .ssh/authorized_keys;
cat tmp.pubkey >>.ssh/authorized_keys;
rm tmp.pubkey"
Make sure to replace <REMOTE> with the hostname you are trying to connect to. Also change id_rsa.pub to whatever your pub key is called.
Cases covered by this:
.ssh directoryauthorized_keys file might not existauthorized_keys does exist it and you want to replace/update/add your ssh public key, this will work without duplicating the entry. This script replaces an entry if there is already one with the same comment.
ssh <REMOTE> "umask 0077; cat > tmp.pubkey ; (...)
ssh-copy-iddoesn't require the private part of the key. All it requires is the public part and the ability to log into the target host (which may require that you type your password or otherwise authenticate). What makes you think that the private key is required? This looks like an XY problem.ssh-copy-id(as of OpenSSH 6.0) doesn't take care of not creating duplicate entries, so you could be unhappy about it for this reason.