Adding Browser Automation to CLI-Anything: First MCP Backend Pattern

Browser automation is genuinely interesting. AI agents navigating websites, filling forms - it's not just nice to have anymore. I recently contributed browser automation support to CLI-Anything via DOMShell's MCP server. Here's how it went. From Issue #90 to PR #118 On March 16, @apireno opened #90 with a compelling idea: "CLI wrappers make software agent-native" - and this should work for browsers too. They mentioned DOMShell, which maps Chrome's Accessibility Tree to a virtual filesystem. Agents control the browser with ls, cd, grep, click - shell commands, not DOM queries. The benchmark was solid: 50% fewer API calls than screenshot-based browsing in Claude tests. What is the Accessibility Tree? The Accessibility Tree is what the browser derives from DOM for screen readers. It's simpler, semantic. Elements are classified by role - button, link, textbox. More stable than DOM, more understandable for agents. DOMShell's insight: filesystem primitives beat DOM queries. ls and grep inste