PHP_CodeSniffer - as you already say - works with the source code at a token level. This is necessary, because it looks at the precise formatting of the code (like whitespace usage).

The parser is more for analyzers that are not interested in the precise formatting of the code, but want to look at the code from a higher level perspective.

For example, if you want to do control flow analysis and type inference working directly on the tokens would be really hard. An Abstract Syntax Tree makes this kind of work much easier, as you don't have to think about the tiny details of the language.