Versions:
AI-Hands, published by AIWander and currently at version 1.0.1, is a single-binary MCP (Model Context Protocol) server written in Rust, designed for agentic browser and desktop automation. The software exposes three primitive subsystems plus a meta-orchestration layer: browser automation via chromiumoxide over the Chrome DevTools Protocol, with the ability to attach to an existing Chrome instance and an accessibility-first approach; Windows UI Automation through the native UIA framework; and vision capabilities combining screenshots with OCR via Tesseract, with PaddleOCR support planned. On top of these primitives, a set of meta-tools orchestrates higher-level workflows, including hands_click, hands_navigate, hands_find, hands_verify, hands_self_record, hands_verify_expectations, hands_summarize_run, and vision_cache, enabling AI agents to perform, verify, and summarize automation runs across browser and desktop environments. AI-Hands fits within the developer tools and AI automation category, specifically targeting MCP server and agentic workflow tooling. It is lean by design, consuming 184 MB per session compared to Playwright MCP's 320 MB, and follows a BYOM (bring-your-own-model) model, working with Claude, Codex, Gemini, GPT, or any MCP-compatible client, making it suitable for users who want to connect their preferred language model to browser control, Windows application interaction, and vision-based verification. Use cases include automated web navigation and interaction, desktop UI testing and manipulation on Windows, OCR-driven screen understanding, and recorded, verifiable agent sessions. Only one version is currently available. After installation, users run installers/scripts/register-hands.ps1 to wire AI-Hands into Claude Desktop's MCP configuration, with the registration process performing a backup-first to preserve existing settings.
Tags: