In general you're better off using the "new" Jupyter kernel for Cadabra which is built by default. On Linux, this is a lot easier than using the old Xeus-based kernel in a conda setup. I will soon remove that support from the source tree altogether, as conda is more and more problematic and there is nobody maintaining those packages anymore.
On Windows, you really are much much better off by running the jupyter notebook server inside WSL, so you can rely on the Linux packages. Then connect to that via a browser on the Windows side.