First, if the error is relatively large and unexpected (assuming you have a good quantitative idea of what to expect) to "calibrate" (ie. adjust out) the error is not a good idea. For many causes it can reappear when conditions change.
As to troubleshooting, I would first carefully read the manual of whatever USB thing you have and see what it says about common mode voltage, grounding and such like. If it says the wrong things I would toss it out and get a better one. It should also give you an idea of the kind of error you can expect. Don't forget to add up all the sources of error in the manual to get the total error. It's common practice to mislead by stating something like repeatability or resolution of a very small figure and then give errors in degrees/degree of cold junction error and a percent of full scale gain error leaving the user to do some work to figure out it's only good for 5 degrees C accuracy and not 0.1 degree C.
Next step would be to attach a proper industrial grade battery powered T/C calibrator to the device itself and make sure it is working properly. If it isn't replace or have the device serviced.
Then I would check the wiring to make sure the correct extension wire has been used and that is connected properly at all junction point. Make sure the thermocouple is the correct type for the instrumentation and the extension wire. Leadwire double reversals are a common source of relatively large errors. Using a J thermocouple with a K instrument or vice versa will result in errors that are (roughly speaking) about 20% of the difference from room temperature, so a 300 degree C reading might be in error by 40-55 degrees.
At that point you can connect the calibrator to the remote point, just to double check, and use the calibrator to measure the thermocouple itself to ensure it's doing what you expect.
This assumes there are no large errors due to junction grounding (eg. shorting o out one leg of the thermocouple), EMI or common mode AC or DC voltage. Such errors are uncommon in properly designed (and expensive) thermocouple instrumentation, but if you have something else there may be an issue. Especially if there is no galvanic isolation.
Bead thermocouples are cheap but often quite inaccurate due to heat loss down the wires. If you can heat sink the wires to the item being measured and ensure good contact between the bead (or junction) and the point you want to measure you will get better results.