Would appreciate peoples opinions on this.

Our tools are often limited by lack of view port interaction. Passing transforms could solve a lot of these cases as we could essentially pass a “rig” to Houdini to assemble. In my perspective a point cloud representation of Maya locator inputs with xform matrices and name as point attributes would be a good solution.

Does people out there have a better one?

There's actually a really old RFE to support inputting locators. This is a really popular requests.
I can't promise anything, but let me see if I can squeeze some time out for this before the next release.