mso

Manef Shell OS

A workspace for your server. A runtime for your agents.

Real terminals, project-aware AI, files and services — in one self-hosted workspace.

Demo · Install · Agent runtime · Use MCP · Docs

See it work

A recorded browser walkthrough, not a mockup. The GIF plays here; click to open it at full size.

Your server, visually	Your tools, in the terminal

More than an AI chat window

Ask it to…	What MSO brings
Understand a project	Project context, trusted skills and task-specific tool discovery.
Do the work	Real PTY, bounded file tools, service controls and explicit approvals.
Pick up where you left off	Durable sessions, local memory, workflow evidence and agent handoffs.
Build server-native workflows	Server-native n8n-core workflow parity: schedule/webhook triggers, branching/loops/subflows, retries/error paths, versions/history, private variables, integrations, per-node logs, and automatic private learning.

One runtime, three ways in: use desktop/mobile windows, stay in your terminal, or connect an MCP client. Code, image/video tools, a browser and native credential setup live beside your operational tools.

flowchart LR
    You["You"] --> Web["Browser · desktop / mobile"]
    You --> CLI["MSO CLI"]
    You --> MCP["MCP client"]
    Web & CLI & MCP --> Core["MSO · auth / scopes / approvals"]
    Core --> Work["Projects · files · PTY · services"]
    Core --> Agents["Agents · skills · durable sessions"]
    Agents --> Evidence["Verify · record · resume"]
    Agents -. BYOK .-> Model["Your model provider"]

Private session screenshots · How the agent runtime works · Architecture · Native Integrations

Install or update MSO from this repo

Linux · normal non-root user · Node 22. One application; no required database or separate agent service. Review scripts/install.sh, then:

curl -fsSL https://raw.githubusercontent.com/rahmanef63/mso/main/scripts/install.sh | bash

mso doctor       # check the installation
mso              # work with the terminal agent
mso web          # open the browser workspace
mso --continue   # resume your last session

The application binds to 127.0.0.1 by default. Use a VPN or protected HTTPS proxy for remote access. Full installation and WSL guide · Android / Termux one-paste install · CLI reference

Update, reset or uninstall — preview before changing anything

mso update updates from main. mso reset and mso reset --scope all preview configuration/factory resets; mso uninstall --purge --remove-code previews removal of owned data and a clean standalone clone. Applying reset/uninstall requires an offline runtime and an exact confirmation token from an independent terminal. Browser reset is separate in Settings → About. Backups, scope and safeguards.

Use MSO from an AI app (MCP)

Connect ChatGPT, Codex, Claude Code, Cursor, or another compatible MCP client to your MSO installation. You do not need SI-Coder: it is an optional MSO plugin, not the MCP server or an installation prerequisite.

1. Open Settings → MCP → Access MSO → Connect an app and choose your client.
2. Copy the server URL shown there (https://mso.example.com/mcp is a placeholder), select OAuth, and authorize on an approved MSO owner device.
3. Grant the minimum useful scope, refresh/scan the client's tools, and enable MSO in the conversation. Start with: @MSO check server status without changing anything.

Direction	What it does
Access MSO	Your AI app → MSO → permitted server/project operations.
MSO Access	MSO → external services/project MCPs with separately configured credentials.

Fresh installs enable MCP with an exec ceiling; existing installs preserve their settings. read observes, write adds bounded changes, and exec permits host/delegated execution. A ceiling is not a grant: actual token scope and server guards still apply. Lower the ceiling when shell access is unnecessary. Never paste passwords or tokens into a prompt.

How to use MCP — Bahasa Indonesia · ChatGPT setup/reference · Protocol and security · Generated tool catalog

Build with it

bun install --frozen-lockfile
bun run verify
bun run test:features
bun run audit:strict

Use Bun >=1.2.15 with native audit support. Contributing · Development · Changelog

Powerful by design. Not a sandbox.

Public Alpha / Developer Preview. Owner/exec authority can run commands as the service user. Provider calls may send selected context off-host. Review approvals and keep credentials private. Successful analysis jobs are not proof of zero open alerts; the security badge above checks the actual inventory. Security policy · Verification and known limits

Product comparison, evidence and limitations · reviewed 2026-08-29.

MIT license · Detailed workspace guide