Claude Code can read my files and run my shell, but out of the box it can't actually go get a page from the live web in a way that survives modern anti-bot defenses. A curl from the Bash tool gets you a 403 from anything behind Cloudflare. So I took an existing open-source scraper — D4Vinci/Scrapling — installed it locally, and registered its built-in MCP server with Claude Code. Now the agent has ten tools for pulling the real web: plain fetches, headless-browser fetches, stealth fetches ...