In this talk, I argue that the extended projection of N (xNP) is a phase (e.g., Bošković 2005) that does not permit successive-cyclic movement via an edge position (cf. Bach and Horn 1976, Bosque and Gallego 2014). As has been observed previously, languages in which AP and xNP can be extracted from xNP (including Chamorro, Hungarian and Russian) typically involve an overt agreement relation between the extractee and the ‘host’ (e.g., Duguine 2008). I argue that in Hungarian-type languages, the agreeing morpheme is itself theta-marked by the host N (cf. Jelinek 1984) and furthermore establishes an interpretable Agree relation with the extractee. Given well-motivated assumptions about adjunction and Agree (cf. Holmberg & Platzack 1995, Brody 1997, Bošković 2007, Hornstein & Nunes 2008, Franco et al. 2015), this enables the extractee to be base-generated outside the host’s extended projection, and hence to be extracted without violating the Phase Impenetrability Condition. As well as accounting for the robust cross-linguistic correlation between overt agreement and extraction, not accounted for under successive-cyclic analyses (e.g., Szabolcsi 1983/84, Gavruseva 2000, Bošković 2005, Ticio 2005, among many others), the proposed analysis accounts for the peripherality restriction on extraction and the possibility of ‘deep’ extraction (e.g., Bošković 2005), as well as certain exceptions to these. Finally, I examine an apparent exception to the agreement/extraction generalisation, the mobility of PP and inherent-case dependents of N, arguing that this can be captured in terms of an Agree relation between the preposition and the head N.