We propose a new kernel for strings which borrows ideas and techniques from information theory and data compression. This kernel can be used in combination with any kernel method, in particular Support Vector Machines for string classification, with notable applications in proteomics. By using a Bayesian averaging framework with conjugate priors on a class of Markovian models known as probabilistic suffix trees or context-trees, we compute the value of this kernel in linear time and space while only using the information contained in the spectrum of the considered strings. This is ensured through an adaptation of a compression method known as the context-tree weighting algorithm. Encouraging classification results are reported on a standard protein homology detection experiment, showing that the context-tree kernel performs well with respect to other state-of-the-art methods while using no biological prior knowledge.