The study evaluates the Indian summer monsoon prediction skill of the Atmospheric General Circulation Model (AGCM) and the impact of sea surface temperature (SST) boundary forcing on the model performance. The National Center for Environmental Prediction's (NCEP's) T170/L42 AGCM model configured with a horizontal resolution of 75 × 75 km, with 42 vertical levels is used for the study. The SST-rainfall relationship is examined in the coupled Climate Forecast System version 2 (CFSv2) model, as CFSv2-predicted SST is used as input for the T170 model. The NCEP Global Forecast System-T170 (GFS-T170) simulations are carried out with boundary forcing of observed SST, CFSv2-predicted SST and the bias-corrected CFSv2 SST. An ensemble of seasonal runs was made using the initial conditions of May to September, and integrated up to September 30th. The significance of discontinuity in the initial conditions due to climate forecast system reanalysis (CFSR) is assessed based on the two-period approach of climatology for the two time scales of 1985–1998 and 1998–2009. CFSv2 predicted climatological summer monsoon rainfall with a significant dry bias over the three convection zones; Western Ghats, Central India and North-east India, and cold bias over the Indian ocean basin and central equatorial Pacific, with strong cold bias over a narrow region of equatorial Pacific. The model could capture 64% (16 out of 25) of the year’s rainfall anomaly signal. The skill of the model is improved in the recent period (1999–2009). The model could simulate the negative Nino 3 and excess rainfall and the La Nina event realistically for the year 1988. The model shows a large difference in Nino indices for the years 1987 and 1998, which led to the unrealistic rainfall simulation. The model has a low skill for indicating the relationship between the Indian Ocean Dipole (IOD) and Indian summer monsoon rainfall (ISMR). The CFSv2 model could not capture the strong positive correlation of the IOD and strong negative correlation of Nino 3 with the ISMR for the period 1999–2009 realistically, suggesting improvement of SST simulation in the CFSv2 model. The T170 model forced with observed SST shows wet bias in peninsular India and dry bias over North-east India, whereas that of CFSv2-predicted SST simulated a wet bias in peninsular India and widespread dry bias in North and Central India. When the model was forced with bias-corrected CFSv2 SST, the dry bias improved in North and Central India, and the intensity of wet bias increased in peninsular India. The model could capture 56, 48 and 64% of the year’s rainfall anomaly signal (positive or negative) correctly in the same sign for being forced with observed SST, CFSv2-predicted SST, and bias-corrected CFSv2 SST, respectively.