How to use t.ppf()? which are the arguments?

According to the reference docs, the arguments to t.ppf are q, df, loc, and scale. The df argument is degrees of freedom, which is usually the sample size minus 1 for a single population sampling problem. Since ppf calculates the inverse cumulative distribution function, by definition a result of x for a given q-value and df means P{T <= x} = q, i.e, there is probability q of getting outcomes less than or equal to x from a T distribution with the given loc and scale. The loc (mean) and scale (standard deviation) arguments are optional, and default to 0 and 1, respectively.

To get a 95% margin of error, you want 5% of the probability to be in the tails of the distribution. This is usually done symmetrically so that 2.5% is in each tail, so you would use q values of 0.025 and 0.975 for the lower and upper cutoff points respectively. For your particular problem, the code would look something like:

from scipy.stats import t

n = 851
mean = 100
std_dev = 0.39

lower_cutoff = t.ppf(0.025, n - 1, loc = mean, scale = std_dev)  # =>  99.23452406698323
upper_cutoff = t.ppf(0.975, n - 1, loc = mean, scale = std_dev)  # => 100.76547593301677