SanctumOS

The Modular, Self-Hosted Agentic Operating System

Sanctum MCP — Orchestration & Control Layer

The Sanctum Model Context Protocol (SMCP) server is the control layer of SanctumOS. It securely exposes tools, scripts, and services to your agents in a way that is modular, auditable, and agent-controllable. Where Broca is the “speech center,” SMCP is the executive function — deciding who can do what, when, and under which rules.

What SMCP Does

  • Governance & Permissions — SMCP enforces who can invoke a tool and under what circumstances, ensuring agents operate safely inside defined boundaries.
  • Orchestration — It coordinates scheduling, retries, and higher-level workflows that tie multiple tools and sub-agents together.
  • Secure Exposure of Tools — From shell commands to DevOps scripts, SMCP makes local capabilities available over a safe, structured protocol.

Key Features

  • Server-Sent Events (SSE) for real-time communication
  • Dynamic Plugin Discovery — All plugins in mcp/plugins/ are auto-registered at startup, no code changes required
  • Automation Support — Built-in helpers for Telegram BotFather flows, CI/CD operations, and deployment rollbacks
  • Single-Thread Safety — Executes plugin actions serially to avoid race conditions, with timeouts and test coverage included

Where It Fits

SMCP is to Sanctum what a kernel is to an operating system:

  • Agents decide what needs to happen.
  • Broca expresses those decisions outward.
  • SMCP authorizes and executes the actions safely.

This division of labor makes SanctumOS not just modular but governable — users remain in control while still giving agents powerful abilities.

Developer Experience

  • Docker-first deployment with aiohttp and pydantic, production-ready
  • Clear docs: Getting Started, API reference, security guides, monitoring, and plugin development tutorials
  • Machine-controllable: every admin action (like flushing a queue or updating settings) can be scripted or invoked by other agents

Why It Matters

Without a governance layer, tool use by agents can become chaotic or unsafe. SMCP solves this by introducing provenance and accountability. Every invocation is trackable, every plugin is sandboxed, and every agent runs under explicit human-set rules.

In short: Broca helps agents talk. SMCP ensures they act responsibly.