How is a PRP different from PRF? Both can be inverted to get the same input

PRP is said to be a bijective function which means that there is a one-to-one mapping with the output. And hence the output can be inverted using the decryption algorithm to get the same input.

Question is how's PRF different from that. If it is not bijective, how are we going to decrypt the message?

In counter mode it is said that we are using a PRF and not PRP (unlike CBC) i.e. we are not using the decryption capabilities of block cipher. What does this mean? We are anyway going to decrypt the message by XORing the IV counter and the key with the cipher text. Then why is it said that we are not using the decryption capabilities? and how is it different from PRP that has to have a decryption algorithm?

Indeed as you noted the primary difference between a PRP and a PRF is that a PRP must be bijective whereas a PRF doesn't have to. This also means that PRFs admit collisions, that is there may exist a key $$k$$ such that for two different inputs $$x,x'$$ $$F_k(x)=F_k(x')$$, whereas this is guaranteed to be impossible for PRPs.