AI-Hands

AIWander·AIWander.AI-Hands

Single-binary MCP server for browser + Windows UI + vision/OCR automation

AI-Hands is a single Rust binary MCP (Model Context Protocol) server that exposes three primitive subsystems plus a meta-orchestration layer for agentic browser and desktop automation: - Browser via chromiumoxide CDP (attach to existing Chrome, accessibility-first) - Windows UI Automation via native UIA - Vision (screenshot + OCR via Tesseract; PaddleOCR planned) - Meta-tools: hands_click, hands_navigate, hands_find, hands_verify, hands_self_record (Item 8+), hands_verify_expectations (Item 13), hands_summarize_run (Item 14), vision_cache (Item 9) Lean by design: 184 MB per session vs Playwright MCP's 320 MB. BYOM (bring-your-own-model) — works with Claude, Codex, Gemini, GPT, or any MCP-compatible client. Post-install, run installers/scripts/register-hands.ps1 to wire AI-Hands into Claude Desktop's MCP config (with backup-first).