I know this is not a programming stack exchange, but... If you have any interest in programming, this would be a perfect task for a simple program and a great learning experience. It could be done in most any programming language.
You can take your PGN separate the moves, input them into a chess engine and give it a certain amount of time to analyze each move. Parse the evaluation for each move and store it. Then loop through the moves, calculate the difference in evaluation from the previous move and classify it as a "blunder" if the evaluation changes more than a defined threshold (1 pt, 2 pts, 3 pts up to you) or perhaps a defined percentage (You may have to consider when the eval is like +10 that it can vary more wildly and maybe omit those scenarios).
If you did this, you could give it a entire list of games, come back in a while and see a list of possible blunders. Otherwise, I think you are left doing it by hand.