Motivated by the scaling limits of the connected components of the configuration model, we study uniform connected multigraphs with fixed degree sequence $\mathcal{D}$ and with surplus $k$. We call those random graphs $(\mathcal{D},k)$-graphs. We prove that, for every $k\in \mathbb N$, under natural conditions of convergence of the degree sequence, ($\mathcal{D},k)$-graphs converge toward either $(\mathcal{P},k)$-graphs or $(\Theta,k)$-ICRG (inhomogeneous continuum random graphs). We prove similar results for $(\mathcal{P},k)$-graphs and $(\Theta,k)$-ICRG, which have applications to multiplicative graphs. Our approach relies on two algorithms, the cycle-breaking algorithm, and the stick-breaking construction of $\mathcal{D}$-tree that we introduced in a recent paper arXiv:2110.03378. From those algorithms we deduce a biased construction of $(\mathcal{D},k)$-graph, and we prove our results by studying this bias.