If reanalysis data (ERA, ECMWF, CSFRR) is too aggregate for you, try looking for the datasets used to calibrate photovoltaics models (PVGIS, PVWATTS and the like) - they include both direct and diffuse insolation.
Also, follow the citation trail for papers that describe the calibration of the PV models, and the reanalysis models - some of the datasets you are seeking will be somewhere in that trail.
Barry Carter has pointed you at the NOAA ISD-Lite dataset, and I'd like to extend that pointer to the whole Integrated Surface Data from NOAA, which is enormous: 20,000 global stations. Because it's an amalgamation of many different sources, they don't all share the same metadata, but this pdf lists the breadth of the metadata: the insolation variables are on pages 64-77.