We describe how we used a Triple Pattern Fragments (TPF) interface and the Comunica knowledge graph querying framework to enable live SPARQL queries over ConceptNet, one of largest knowledge graphs for commonsense reasoning publicly available on the Web. Despite being a Linked Data resource, the official ConceptNet is not published in RDF and does not support SPARQL. Instead, it provides a RESTful API for live queries, which are restricted to simple triple patterns. This limited API makes it hard for users to search for non trivial patterns in the graph and hinders the possibility of federated queries offered by SPARQL. There have been attempts to convert ConceptNet to RDF but such proposals tend to quickly become obsolete. In this paper, we take a different route. We use TPF to expose a low-level RDF query interface to ConceptNet. This low-level interface is built on top of the ConceptNet API and can be used by TPF-compatible SPARQL engines such as Comunica. Using this approach, we were able evaluate non-trivial SPARQL queries, including federated queries, over ConceptNet on-the-fly. Our experiments showed that overhead incurred is small and can be further reduced by optimizing ConceptNet’s internal edge representation. We argue that such overhead is justified by the gains in expressivity and flexibility. Moreover, the overall approach is general and can be extended to other non-RDF knowledge graphs.