Here's a counterexample. Let $k$ be a field and take $A=B=k[x]/(x^2)$ and $M=C=k$ (considered as an $A$-algebra via $x\mapsto 0$). Then $\mathrm{Hom}_B(M,B)\cong k$ (spanned by $1\mapsto x$), so $\mathrm{Hom}_B(M,B) \otimes_A C\cong k$ as well. But $1\mapsto x$ maps to $0$ in $\mathrm{Hom}_B(M,B \otimes_A C)$, since $x=0$ in $B\otimes_A C=C$.
On the other hand, if you assume $C$ is flat, then the map always is injective (in fact, it is an isomorphism). Indeed, you can note that there is a natural isomorphism $\operatorname{Hom}_B(M,B)\otimes_A C\cong \operatorname{Hom}_B(M,B)\otimes_B (B\otimes_A C)$ that is compatible with the canonical maps to $\mathrm{Hom}_B(M,B \otimes_A C)\cong \mathrm{Hom}_B(M,B\otimes_B (B \otimes_A C))$, so you are reduced to this question over the ring $B$ with $N=B$ and $P=B\otimes_A C$ (which is flat over $B$ since $C$ is flat over $A$).