Bad tcp

How did you test it? By any chance is your robot equipped with the absolute accuracy package? If you tested while jogging try again with a program. While jogging absolute accuracy model is turned off.

Fubini

things can also get really weird when wrong MADA is used or if robot is not mastered correctly or if tool is not mounted correctly (without dowel pin) or if tool is flexible or if reference point moved during measurement...

Quote from MIJ

.. when I tested the tcp I found that it's way off...

What does this mean? Quantify 'way off'

you still did not answer the question. "way off" and "all over the place" are not a number.
quantify it means measure it, provide a value... 1mm, 10mm, 100mm, 1km...

if the wobble is small (~1-3mm) it may be normal for the robot and chosen TCP.
if the wobble is larger (~ 3-10mm) it is likely that reference point moved or tool shifted.
if the wobble is even larger it is likely that problem is with mastering or MADA or belts slipping etc.

in fact you did not provide any data. what is the measured TCP you got? is it far from flange?

Quote from MIJ

Means when I rotate the tool around A,B, C it's going all over the place instead of holding on to the center.

Yes, but how much? How many mm of displacement per degree of rotation? On which axes? Quantifying the characteristics of the error is important. Orientation changes are the worst case for robot accuracy -- even a perfect TCP can easily drift 1mm over 90deg of rotation.

Usually, if the MADA are incorrect, achieving a low error on the TCP 3-point process isn't possible. Incorrect LOAD_DATA could be a potential cause of what you're seeing. Activating the wrong Tool could be another. As has been stated upthread, the best test is to create a test program that touches the TCP to a reference point at multiple rotations, rather than jogging the robot manually.