A cluster is simply defined by a property file having property indexclass equal to the suitable cluster class (e.g., it.unimi.di.big.mg4j.index.cluster.DocumentalConcatenatedCluster), a clustering strategy specified by the property strategy and one localindex property for each local index (see also the documentation of it.unimi.di.big.mg4j.index.cluster.IndexCluster). You can run IndexBuilder with the --keep-batches option to have a look at the generated cluster files, which expose the batches as a single index.