mgc-blackbox 1.3.0

A trusted boundary for intent‑safe AI execution.

MirginCipher Blackbox (MGC)

MirginCipher Blackbox (MGC) is a Local Encrypted Execution Tool designed to protect sensitive human intent and enable secure, deterministic AI execution. MGC provides a trusted device‑level encrypted boundary for agents — it is not an agent itself.

Authorization Integration into any third‑party products or AI agents is free of charge, but requires official authorization to ensure ecosystem integrity and security.

For authorization requests, please contact: [e-mail]

Core Capabilities

• Local storage — All sensitive data stays on the device and is never uploaded to the cloud.
• Local encrypted storage & execution — Data is encrypted using a combination of the user’s personal key and device‑specific environmental factors.
• Store‑once authorization — Once stored, an item is authorized for future use within the same device environment without repeated confirmation.
• Environment migration — If hardware changes, access can be restored using a user‑defined migration key. Stored data remains usable.
• Lightweight integration — Install the MGC package and configure the dedicated MGC Skill. Even if Skill‑level communication is intercepted or modified, MGC’s secure storage and retrieval remain unaffected.
• Cross‑agent availability — Any agent platform that supports Skills can integrate with MGC with zero additional development.
• Cross‑platform support — Distributed as a Python package for easy installation, with all security‑critical components compiled via Cython and executed inside a local encrypted service.This design ensures portability, tamper‑resistance, and consistent behavior across macOS, Windows, and Linux.

Installation pip install mgc-blackbox

Start the local encrypted execution service: mgc run

Usage Overview MGC can be used in two ways:

1. Through AI agents (via Skills / MCP) Agents can:

• Store sensitive information
• Retrieve encrypted items
• Execute stored scripts
• All without accessing plaintext.

MCP configuration: See mcp_config.json in installation directory.

2. Through scripts (via REST API) External scripts can fetch encrypted items at runtime. Plaintext is never exposed to AI logs or external systems.

For detailed usage, see: MGC_GUIDE.md

Security Model All data remains local No cloud upload No plaintext logging Deterministic execution User‑controlled authorization Protection Mode available for high‑security environments Minimal network usage: only version checks and service health checks require network access;no user data, secrets, scripts, or metadata are ever transmitted.

For safety details, see: docs/user_notice.md

AI Skill Specification For AI behavior boundaries and tool definitions, see: docs/skill_spec.md

License See the LICENSE file for full terms.

© 2026 MirginCipher Team. All rights reserved.