Error on constrained fit parameters using lsqcurvefit

I'm using lsqcurvefit to fit some experimental data to a series of exponentials. Other fitting routines (fitnlm, etc) do not perform as well, and a number of parameters must also be constrained to fit properly. To get the uncertainty on each parameter, I am using nlparci with the Jacobian output of lsqcurvefit. However, I notice that the errors are quite large in a lot of cases, compared to when I leave the fit unconstrained (even though the fit is worse). I have seen conflicting information about whether nlparci is valid when the fit is constrained. My question is, can I use nlparci to find parameter errors when using constraints on lsqcurvefit? If not, how can I find these errors? Thank you.