Jharokha

•

Enables AI agents to view and optionally interact with Windows remote desktop sessions via local OCR and guarded input controls.

Local MCP server that gives AI agents eyes, and optionally hands, on a Windows remote-desktop session. Reads the screen as text; clicks, types and scrolls only when you allow it.

主な機能

010 GitHub stars
02Non-intrusive local OCR extraction that converts remote Windows UI into structured text without requiring session-side software
03Safe-by-default architecture enforcing read-only operation unless explicitly started with interactive input permissions
04App-aware input guardrails that prevent typing into wrong apps, sending accidental messages, or interfering with active user input
05Predefined high-level workflows for common apps including Teams messages, Outlook emails, browser interactions, and terminal commands
06Privacy-first design with local-only processing, zero network transmission, and no telemetry

ユースケース

01Allowing Claude or other AI agents to triage and search emails or Teams messages in secure corporate virtual desktops
02Executing guarded terminal commands and file transfers across remote desktop environments
03Automating navigation and data extraction in legacy Windows remote apps where standard APIs are unavailable