Right now, the server just looks in the working directory - which in the current typical use case (a Windows binary snapshot) is the bin directory. Instead of loading JSON files directly, it's reasonable that we might consider having a module in charge of loading json files that would handle paths, including per-user config paths (XDG standards on Linux, AppData on Windows, elsewhere on other platform) as well as varying system-wide paths (Linux should get config files in /etc, Windows you've got the program directory itself, ProgramData for system-wide stuff)
h/t @feilen for bringing this up and prompting filing it as an issue in GitHub rather than my brain.