Maybe more documentation should be added, like instructing the users to setup the solver and some solver-specific parameters?

A good example is the contribution of Xpress support: #62

I will work on that. Thanks @metab0t.

I checked and add KNITRO where I think is missing. Please let me know why if anything is missing.

Some suggested additions:

Callback (I am not sure how KNITRO callbacks work):
https://github.com/metab0t/PyOptInterface/pull/62/changes#diff-c1197ba52e5b09376138ed8bcafa99afdee21a5ae007bc19c7d0e31ca6b5cd88

Setup of KNITRO optimizer:
https://github.com/metab0t/PyOptInterface/pull/62/changes#diff-5b33c125262c6fe31731295d534d95585911770c494569f19fa9b090c58419ae

License context and solver-specific parameter or attribute:
https://github.com/metab0t/PyOptInterface/pull/62/changes#diff-f1a26b9aa5d360f3bd76d6cb059b535f5c82eb4174fb4f53b7a54e1bfac2f006

Thank you for pointing this out.
Regarding solver-specific attributes and parameters, I included everything I believe is relevant in the initial PR. Unfortunately, KNITRO callbacks are quite complex to handle properly, so I decided to leave them out of this first version of the implementation. I do plan to add support for them in the future; for that reason, I have not mentioned them in the documentation yet.
As for the license context manager, I am currently working on it and will add support soon. Once it is implemented, I will update the documentation accordingly.

@metab0t I just updated the doc of KNITRO to include the Env (license manager) documentation. I also mentionned that currently the callbacks are not supported. Please let me know if anything else is needed.

Thanks! LGTM

The Linux build's failing. What do you think's causing it?

Nothing to worry. It occasionally segfaults after testing due to unknown reasons on Linux.