Parsing text wouldn't be the way to go; you can use google services to get search results and format them to your liking (subject to google's terms and conditions). Writing robots to crawl the entire web and a search engine to search the results is an enormously large project. It would really surprise me if it weren't better to use an existing search engine.

For those that haven't seen it, the Binary Search Tree 2 and the report on it are interesting reading.