These descriptions reflect the differences in what Langrange's and Kane's method produces.
Lagrange's method produces a kinematical equation (linear in q'), a dynamical equation (linear in q''), and a constraint equation that includes Langrange multipliers (if there are constraints). The m_c, m_dc, and m_d are simply the matrices that contain the coefficients to the linear terms in the equations. m_dc is often called the mass matrix (simple example: m x'' = -c x' - k x - f ). In Lagrange's method m_c is always identity because you simply sub in u = q' to put the equations in first order form.
Kane's method produces a different, but equivalent, set of equations.
1. holonomic constraint equation (non-linear kinematic loop constraints) [ this can also be in Langrange's method as a 4th equation but we currently don't have it explicitly defined]
2. kinematical differential equations (linear in the generalized speeds). This is typically u = q' like in Lagrange's method but Kane's method allows you to define these any way you want so that you can get simpler equations in the end.
3. non-holonomic constraints (linear in the generalized speeds and defines the relationships between the independent speeds and the dependent speeds)
4. dynamical equations of motion (linear in independent speeds)
5. dynamical equations of motion (linear in dependent generalized speeds)
3, 4, and 5 serve to describe the same thing that the last two Lagrange equations describe.
The k terms are simply the linear coefficient matrices in those equations.
Theoretically you should be able to transform the results of Lagrange's method to those of Kane's. I'm not sure if that procedure is laid out anywhere in the literature. But it may have some complications.
The f's are simply vector equations of those variables that are non-linear. They will hold all external forces, but also terms that are non-linear like the Coriolis forces, etc.