Why is scaling for numeric variables done before splitting it into a train & test dataset?

Why do we need an intercept in a linear regression model? If we use statsmodel.OLS do we need to add the intercept explicitly and how do we do it?