I made Claude's web research 18× cheaper with 2 lines of setup

I've been using Claude Code daily, and one thing kept bothering me. Whenever Claude calls WebFetch , it often dumps 3,000–15,000 tokens from an entire web page into the context window, even if the answer is buried in a single section. Multiply that across a few documentation pages, and you're spending thousands of unnecessary tokens just to answer one question. So I built Webify . Instead of sending the whole page to Claude, Webify parses the HTML into a DOM graph, identifies the parts relevant to your quer