The Cheugugi rain-gauge record of Seoul, Republic of Korea (1777–1907), is among the world's longest instrumental daily precipitation records and a rare quantitative gauge of the pre-industrial East Asian monsoon. Here, we reconcile it against the original chronicle and use it to document, and attribute, a severe sustained rainy-season drought from 1881 to 1907: rainy-season (June–August, JJA) precipitation fell ~27 % and annual totals 28 % (with a Pettitt change point in 1880, <em>p</em> ≈ 0.01), with the contribution of days >100 mm falling by ~43 %. Three distinct record channels – the gauge, a separate documentary reconstruction, and qualitative rain reports in the Seungjeongwon Ilgi – registered the same shift, arguing against any decay in observational practice. In the Twentieth Century Reanalysis, the unconstrained precipitation field did not resolve the drought, whereas its pressure-constrained circulation registered a weakened monsoon low. Pacemaker experiments (10 members each) were conducted to identify the oceanic driver, prescribing the observed sea-surface-temperature anomalies over the tropical Pacific, North Atlantic, and Indian Ocean individually and globally. The historical forcing produced a significant weakening of the low-level monsoon circulation over Korea: a nearly 8 m depression of the JJA 850 hPa geopotential height (<em>p</em> < 0.001, the most robust signal), weakened southerly inflow, and significant reduction in the vertically integrated moisture supply. The geopotential-height weakening was driven principally by the Indian Ocean, the largest single-basin contributor, together with the North Atlantic, and not by the tropical Pacific despite its El Niño-like state. The reconciled record supplies a pre-industrial monsoon baseline and, with the experiments, attributes the monsoon-circulation weakening that accompanied the historical Korean drought to Indian Ocean and North Atlantic cooling. Future work should disentangle the tropical-cyclone contribution to the heavy-rainfall deficit and confirm the attribution with additional models.