We present the first robust measurement of the high redshift mass-metallicity (MZ) relation at 10(8) less than or similar to M/M-circle dot less than or similar to 10(10), obtained by stacking spectra of 83 emission-line galaxies with secure redshifts between 1.3 less than or similar to z less than or similar to 2.3. For these redshifts, infrared grism spectroscopy with the Hubble Space Telescope Wide Field Camera 3 is sensitive to the R-23 metallicity diagnostic: ([OII] lambda lambda 3726, 3729 + [OIII] lambda lambda 4959, 5007)/H beta. Using spectra stacked in four mass quartiles, we find a MZ relation that declines significantly with decreasing mass, extending from 12+log(O/H) = 8.8 at M = 10(9.8) M-circle dot, to 12+log(O/H)= 8.2 at M = 10(8.2)M(circle dot). After correcting for systematic offsets between metallicity indicators, we compare our MZ relation to measurements from the stacked spectra of galaxies with M greater than or similar to 10(9.5)M(circle dot) and z similar to 2.3. Within the statistical uncertainties, our MZ relation agrees with the z similar to 2.3 result, particularly since our somewhat higher metallicities (by around 0.1 dex) are qualitatively consistent with the lower mean redshift (z = 1.76) of our sample. For the masses probed by our data, the MZ relation shows a steep slope which is suggestive of feedback from energy-driven winds, and a cosmological downsizing evolution where high mass galaxies reach the local MZ relation at earlier times. In addition, we show that our sample falls on an extrapolation of the star-forming main sequence (the SFR-M-* relation) at this redshift. This result indicates that grism emission-line selected samples do not have preferentially high star formation rates (SFRs). Finally, we report no evidence for evolution of the mass-metallicity-SFR plane; our stack-averaged measurements show excellent agreement with the local relation.