As Mike says, it's locally finitely presentable because $Cat$ is (equivalent to) the category of models of a finite limit sketch. A quick way of describing this is to say that a category is (or is the nerve of) a simplicial set such that certain squares in the combinatorial definition of simplicial set (via faces, degeneracies) have to be pullbacks, as stated more precisely in the sixth proposition of this section of the nLab article on nerves. This clearly gives the notion of category in terms of a finite limit sketch. (In fact, this will work with truncated simplicial sets $C$, involving only $C_j$ up to dimension $j = 3$ if I'm not mistaken.)
In other language, every category is a filtered colimit of finitely presented categories (meaning categories presented as coequalizers of pairs
$$R \rightrightarrows S$$
where $R, S$ are free categories on finite graphs). Colimits in $Cat$ can be a little bit unpleasant; coproducts are of course unproblematic, but coequalizers are another matter. An explicit construction of coequalizers can be found in this paper. (I mean to come back to this answer and improve it if possible.)