MaxEnt can - cautiously - be used for predicting future suitable habitat. For a species to actually occupy this new habitat requires that it can get there, along with any essential mutualists, and outcompete the species that currently occupy the area. SDMs correlate current spatial distributions with current climatic and other (soil, topography etc.) variables and then predict the future distribution from the future climate. Such correlative models are easy to run and are generally fairly accurate in reproducing the current distributions of plants. However, they have a number of limitations when used for prediction, including the inherent dangers of extrapolation from correlations and their inability to deal with novel future climates (or novel combinations of climate and other factors). If you use only climatic variables in your model, then you have to assume that climate entirely determines the current distribution, which is unlikely.

@ Corlett

I have recommended your response.

Nonetheless, the response contains statements that may confuse novices in this field. The Maxent algorithm does neither notice whether the climate, nor the presence points are "current". Neither is the algorithm specific for a species. Any object with an unbiased set of point locations in geographical space will do.

Indeed, using only (bio-)climatic variables, typical for the pioneers of maxent use in ecology, may be suitable for large areas across latitudes (somehow with a preference for South America). Even in such cases, classical meteorologic variables possibly outperform the bio-climatic ones. For prediction in smaller areas and/or at finer resolutions, all fundamental categories of environmental predictor variables (indeed, soil, topography etc) may be required. Hierarchy theory, in other words. I rest my case with reference to our publications.

Using Maximum Entropy Modeling (MaxEnt) to predict future trends in the distribution of high altitude endemic insects in response to climate change --is a good article