Abstract Many intrusions with adakite-like affinities in collisional zones have obviously higher K2O contents and K2O/Na2O ratios compared with counterparts in subduction zones. A suite of Eocene post-collisional high-K2O adakite-like intrusions, mafic microgranular enclaves, and potassic–ultrapotassic lamprophyres in the Machangqing complex are associated with the Indian–Asian collision within the western Yangtze Craton, southeastern Tibet. The potassic–ultrapotassic lamprophyres, with a zircon U–Pb age of 34·1 ± 0·2 Ma, have high K2O and MgO contents, are enriched in light rare earth elements and large ion lithophile elements, and display high Rb/Sr, and low Ba/Rb and Nb/U ratios. They show enriched isotopic compositions [i.e. (87Sr/86Sr)i = 0·7070–0·7082, εNd(t) = −3·2 to −2·8], and zircon εHf(t) values (−1·6 to +2·6). Their parental magmas are inferred to have been derived from partial melting of an enriched lithospheric mantle, metasomatized by subduction-related fluids. The adakite-like intrusions, with zircon U–Pb ages of 35·4 ± 0·4 and 35·2 ± 0·3 Ma, are characterized by high SiO2 (68·8–71·1 wt%) and Al2O3 (14·0–15·3 wt%) contents, high Sr/Y (41–118) ratios, and low Y (5·3–14·7 ppm) contents. They show low contents of compatible elements (e.g. Ni = 9·5–36·2 ppm) and total REE, and lower Mg# values than the lamprophyres and mafic microgranular enclaves. The adakite-like intrusions have positive large ion lithophile element anomalies, especially potassium, negative high field strength element anomalies, negative εNd(t) (−5·5 to −3·3), and high (87Sr/86Sr)i (0·7064–0·7070) and zircon εHf(t) values (0·0 to +2·7), indicating that they were formed by partial melting of the juvenile lower crust. Mafic microgranular enclaves hosted in the adakite-like intrusions, with U–Pb ages similar to the lamprophyre of c. 34 Ma, exhibit disequilibrium textures, and some of them contain phlogopite. They exhibit potassic–ultrapotassic affinity, and relatively high compatible element contents. They are also characterized by enriched isotopic compositions with (87Sr/86Sr)i = 0·7063–0·7074, εNd(t) = −6·6 to −4·1, and variable zircon εHf(t) values (−0·6 to +3·2). Petrological and geochemical evidence suggests that the mafic microgranular enclaves were formed by magma mixing between potassic–ultrapotassic and pristine adakite-like melts. We propose a magma mixing model for the origin of the high-K2O adakite-like intrusions from the Machangqing complex. In this model, the formation of high-K2O adakite-like intrusions occurred in three stages: (1) partial melting of metasomatized lithospheric mantle generated potassic–ultrapotassic mafic melts; (2) underplating of these mafic melts beneath thickened juvenile lower crust resulted in partial melting of juvenile mafic lower crust and the generation of adakite-like melts; (3) magma mixing involved 80 % pristine adakite-like melts and 20 % potassic–ultrapotassic melts. This leads to the enrichment of K2O in these adakite-like intrusions, and magma differentiation further promotes K2O enrichment. These results are applicable to compositionally similar adakite-like rocks produced in other collisional zones, such as the Tibet, Sulu–Dabie and Zagros orogenic belts. From which we conclude that in continental collision zones, the post-collisional mantle-derived magmas characterized by potassic–ultrapotassic affinities are spatially associated with coeval collision-related adakite-like intrusions that originated from lower crustal melting. The emplacement of adakite-like and potassic–ultrapotassic rocks is controlled by the same fault systems, which increases the possibility of interaction between these two magma suites.