See the answers under screen scraping; basically they load the web pages up, login as you, and parse the html to find what they need...

ironically you can just grep it:)

curl -d "username=foo&password=bar" https://yourbank.com | grep balance