In the minimum-description-length formulation of Occam's razor, we are trying to find the shortest computer program (smallest model) that completely and perfectly predicts all the data we're trying to model. Simplicity is simply how many characters (or bits, etc) the model takes to formally describe.
The smallest model that does this is not necessarily the right one, but it is considered more likely than other models, until and unless some new data contradicts it.
Solipsism concerns the universe in general, so the "data we're trying to model" in this case would be all of our observations of the universe.
The first thing to see is that solipsism, by itself, does not completely and perfectly predict all those observations. The most we can say is that solipsism is not incompatible with them - but it is not a complete theory. It makes no predictions at all about what would be observed.
In order to evaluate solipsism in this setting, we need to view it as a possible part of a larger, complete model that does correctly predict all the observations. The question then becomes: does the presence of solipsism in the model help to reduce the length of the model?
To that, I'd say the answer is no. Adding solipsism to your universe-model would only serve to make the model longer and more complex, without adding anything to the model's predictive capabilities.
For an example of what such a model would look like, the model could say that you are a brain-in-a-vat, and the rest of the universe, including every other person, is a computer simulation fed into the brain, not real. Loosely speaking, you could call this a solipsistic universe-model. Such a model needs to include (A) the vat and the brain and how the brain is hooked up to the simulation, and (B) the details of how the simulation works. But, if we compare it to a non-solipsistic universe-model, the non-solipsistic universe-model only needs to include the equivalent of (B), and can omit (A), so it is shorter and simpler.
You might object that this is not really solipsistic because in addition to the self the model also admits to the reality of a vat, wires, and a computer. Granted; it's just to give you an idea. Any solipsistic universe-model would make a separation between the self and the rest of the universe, assigning the status of "real" to the self and "non-real" to the rest of the universe. This is going to add some complexity compared to a universe-model where that separation is absent and the self is modeled with the same code used for the rest of the universe.