We search for high redshift ($z\sim$1-2) galaxy clusters using low luminosity radio galaxies (FR~I) as beacons and our newly developed Poisson Probability Method (PPM) based on photometric redshift information and galaxy number counts. We use a sample of 32 FR~Is within the Cosmic Evolution Survey (COSMOS) field from Chiaberge et al. (2009) catalog. We derive a reliable subsample of 21 {\it bona fide} Low Luminosity Radio Galaxies (LLRGs) and a subsample of 11 High Luminosity Radio Galaxies (HLRGs), on the basis of photometric redshift information and NRAO VLA Sky Survey (NVSS) radio fluxes. The LLRGs are selected to have 1.4~GHz rest frame luminosities lower than the fiducial FR~I/FR~II divide. This also allows us to estimate the comoving space density of sources with $L_{1.4}\simeq 10^{32.3}\,\hbox{erg}\,\hbox{s}^{-1}\,\hbox{Hz}^{-1}$ at $z\simeq 1.1$, which strengthens the case for a strong cosmological evolution of these sources. In the fields of the LLRGs and HLRGs we find evidence that 14 and 8 of them reside in rich groups or galaxy clusters, respectively. Thus, overdensities are found around $\sim70\%$ of the FR~Is, independently of the considered subsample. This rate is in agreement with the fraction found for low redshift FR~Is and it is significantly higher than that of FR~IIs at all redshifts. Although our method is primarily introduced for the COSMOS survey, it may be applied to both present and future wide field surveys such as SDSS Stripe 82, LSST, and Euclid. Furthermore, cluster candidates found with our method are excellent targets for next generation space telescopes such as JWST.