I'm using Savyasachi Singh's template here to get the pointer of a TypedArray. Then I map this pointer to an Eigen matrix for the heavy computation. The example is like the following. Note that to use external c++ libraries with C++ MEX API, we need to write a wrapper (e.g. compute_something_with_eigen.cpp), and operator() can internally use this function. After we finish the computation with external libraries, we can find the pointer of this result matrix and map it back to a TypedArray.
// Assign the input, assume it's a 2D matrix
TypedArray<double> matrix = std::move(inputs[0]);
// Use the template I mentioned above
double* ptr = getPointer(matrix);
// Get the dimension of input
size_t size_input = matrix.getDimensions();
// Don't do the following in your C++ MEX main file
MatrixXd mat_eig = Map<MatrixXd>(ptr,size_input[0],size_input[1]);
// Do your algorithm
