Context. Planck will provide an unprecedented view of the (cold dust in both diffuse regions and dense cores and associated gas in the Galaxy), resulting in the observation of a large number of cold objects. Understanding the processes acting at the very earliest stages of star formation requires a detailed study of these cold cores. Aims. We present an algorithm which is able to compile a reliable catalogue of cold cores from the Planck High Frequency Instrument (HFI) data. We construct models of the cold cores and validate the extraction algorithm on simulated Planck data. Methods. We model cold cores along with their warmer envelopes which are then injected into a full sky simulation of Planck HFI data. These cold cores are mostly embedded in the relatively warmer Galactic background, in particular in the plane where there is a high level of confusion. We have developed two detection algorithms which make use of the colour of the cores rather than simply extracting compact sources from the individual frequency maps. The cold sources present a relative emission excess at wavelengths longer than 350 μm (i.e. 857 GHz), and thus provide the necessary signature to separate them from the diffuse Galactic emission and warmer cores. Furthermore, using an optimised photometry algorithm for embedded cold sources, reliable fluxes can be extracted. Results. We recover nearly 3% of our input sources, with 60% completeness at high latitude. This suggests that we will be able to compile a catalogue of nearly 5000 objects over the entire sky with a reliability of 98%. These detections cover the entire simulated mass range from 0.4 to 10 000 solar masses and cover the entire sky, including a small number of sources on the far side of the Galaxy. Conclusions. This algorithm is ready to be applied on Planck HFI data, and will be able to produce a reliable catalogue of sources ready for follow up observations, in particular for the Herschel open-time key-program on Galactic cold cores.