On June 4, 2002 CNMAT released version 2.2 of the mlp object, which correctly reads training data files, allowing you to train networks yourself.
CNMAT has a Max object "mlp" (short for "multi-layer perceptron") that implements simple backpropagation learning and forward-pass.
Note that it requires a bit of sophistication to use this object.
CNMAT does not have any plans to revive the graphical features of the old mlp object.