Frankel (section $18.1$a) calculates the Maurer-Cartan form $\theta$ of the Lie group $\operatorname{SO}(2)$ by using the formula $\theta=g^{-1}dg$, where $g$ is viewed as the identity on $\operatorname{SO}(2)$,$$dg=\begin{pmatrix}d(g_{11})&d(g_{12})\\d(g_{21})&d(g_{22})\end{pmatrix}\in \operatorname{Mat}(2\times 2)\otimes\Omega^1(\operatorname{SO}(2))$$ is a matrix of $1$-forms and $\theta=g^{-1}dg$ is simply the product of the two matrices, i.e.$$\theta_{ij}=\sum_k (g^{-1})_{ik}(dg)_{kj}\in\Omega^1(G).$$ In principle, this makes sense for any matrix group, i.e. if $G\subseteq \operatorname{GL}(n)$ is a Lie subgroup, then we can use the last equation to define the Maurer-Cartan form. Of course the question is whether this is equivalent to the standard definition using the pushforward of the left translation. One issue is obvious:
The definition described above yields an element$$\theta\in\operatorname{Mat}(n\times n)\otimes\Omega^1(G)$$and it is not clear that $\theta$ is even an element of the subset $\mathfrak g\otimes\Omega^1(G)$, i.e. $$\theta\in\mathfrak g\otimes\Omega^1(G)\subseteq \operatorname{Mat}(n\times n)\otimes\Omega^1(G).$$ In the example discussed in the beginning this can be checked explicitly, but is there a more general argument?