Thank you for the ideas.
I will switch to API later then, since xidel supports XML (maybe JSON too?). It'll be quite hard for me since API is giving a lot of useless (for what i'm doing) info. I literally only need tags as a plain text, delimited by ",". I'm not even parsing the actual post page, but the search results page tooltips, because they are closer to what i want there. :)
See: http://c2n.me/3nX66OM.png
Parsing that XML page actually look like a lot of more work from the first look. Okay it's actually not that bad after i'll figure out "properties" i think...
But i like the fact that i can export artist tag into appropriate XMP tag. Without tag types, that was impossible.

Regarding filenames, i'm not sure what else can i rely on to automate the process. MD5 sounds like a best shot.
And if the md5 has changed on the server side, there is no way to find a new image automatically AFAIK.

Not sure why would anyone rename his files, but i'll consider adding MD5 checks later.