We present the X-ray pipeline developed for the purpose of the cluster search in the XMM Large-Scale Structure (XMM-LSS) survey. It is based on a two-stage procedure via a dedicated handling of the Poisson nature of the signal: (i) source detection on multiresolution wavelet filtered images and (ii) source analysis by means of a maximum likelihood fit to the photon images. The source detection efficiency and characterization are studied through extensive Monte Carlo simulations. This led us to define two samples of extended sources: the C1 class that is uncontaminated, and the less restrictive C2 class that allows for 50 per cent contamination. The resulting predicted selection function is presented and the comparison to the current XMM-LSS confirmed cluster sample shows very good agreement. We arrive at average predicted source densities of about seven C1 and 12 C2 per deg2, which is higher than any available wide field X-ray survey. We finally note a substantial deviation of the predicted redshift distribution for our samples from the one obtained using the usual assumption of a flux-limited sample.
