Rabbit Commander
Manage devices
RABBIT COMMANDER // COMPLETE PRODUCTION SETUP GUIDE

Install once. Run silently. Control securely from your AI client.

This guide covers Windows 10/11, Linux servers, VPS hosts, RunPod GPU pods, macOS, ChatGPT, Claude, Gemini CLI and compatible remote-MCP clients. The production relay is wss://commander.zippri.com/relay and the preferred MCP endpoint is https://commander.zippri.com/mcp-native-v1.

Security rule: Never paste a Device Token, OAuth access/refresh token, password, API key or private key into an AI chat, screenshot, email or support ticket. Device Tokens are for the agent installer only.
1. Create device2. Windows3. Linux / VPS4. RunPod5. macOS6. ChatGPT7. Claude8. Gemini CLI9. Other MCP10. Troubleshooting
STEP 1

Create your Rabbit Commander device

Every computer, server, VPS or GPU pod gets its own Device ID and one-time Device Token.

CONTROL PLANEPAIROne device at a time

Account and device

1
Sign in

Open https://commander.zippri.com and sign in to your Rabbit Commander account.

2
Create the device

Open Devices -> Add device. Give it a recognizable name such as Office Windows PC, Production VPS or Agastya RunPod.

3
Copy the pairing values

Copy the Device ID and Device Token exactly. Keep the token private. Do not confuse the ID with the token.

4
Install on that exact machine

Use one of the platform sections below. After installation, return to Devices and wait for ONLINE.

Network requirements

Outbound HTTPS/WSSAllow outbound TCP 443 to commander.zippri.com. No inbound port required for the agentThe device initiates the secure WebSocket connection to the Rabbit Commander relay. Stable clockKeep system time synchronized; OAuth and TLS flows depend on accurate time. Proxy/firewallIf your organization intercepts TLS, make sure the machine trusts the corporate CA and allows WebSocket upgrades.
WINDOWS 10 / 11

Windows setup - silent background agent

The recommended EXE bundles the Node.js runtime and installs Rabbit Commander as a persistent scheduled task. The production agent must run in the background without a permanent PowerShell or Windows Terminal window.

INSTALLER1.1.0Bundled Node.js

Recommended: Setup EXE

NO NODE SETUP
Download Windows Setup EXE 1.1.0
1
Run as Administrator

Right-click the downloaded EXE and choose Run as administrator.

2
Paste Device ID

Paste only the Device ID created in Rabbit Commander.

3
Paste Device Token

Paste the private Device Token when prompted. It is stored under C:\ProgramData\RabbitCommander\Agent.

4
Finish and close installer

The agent continues silently in the background. A terminal window should not remain open after setup.

Expected background behavior: Task Scheduler starts a hidden launcher. The agent reconnects automatically after login and restarts after an unexpected exit. Logs remain available without keeping a console window open.

Verify Windows installation

SAFE CHECKS
schtasks /Query /TN "Rabbit Commander Agent" /V /FO LIST Get-ScheduledTask -TaskName "Rabbit Commander Agent" | Select-Object TaskName,State Get-Content "C:\ProgramData\RabbitCommander\Logs\agent.log" -Tail 50 Get-ChildItem "C:\ProgramData\RabbitCommander\Agent"

Return to the Rabbit Commander Devices page and confirm the device shows ONLINE.

If an old blinking terminal is still open

  1. Install the current Windows EXE again; it stops/replaces the previous Rabbit Commander scheduled task. If you no longer have the one-time Device Token, rotate the token from the device controls first and use the newly issued token.
  2. Close the old terminal window that says the process exited with code 0. Do not keep pressing Enter to restart that old session.
  3. Verify the scheduled task is running, then confirm the device is ONLINE.
  4. If a console still appears after the updated install, capture only the window title and task state - never include your Device Token.

ZIP fallback / manual package

Download Windows ZIP 1.1.1
  1. Extract the ZIP to a normal local folder.
  2. Open the windows folder.
  3. Right-click Install-RabbitCommander.cmd and run it as Administrator.
  4. Paste the Device ID and Device Token when asked.
  5. Verify with the commands above.
LINUX / VPS

Linux server and VPS setup

Use the systemd installer on Ubuntu, Debian and other systemd-based hosts. The service starts at boot and restarts automatically.

SERVICEsystemdRestart always

Install

NODE 18+
Download Linux Package node --version npm --version

If Node.js is not installed, install Node.js 18 or newer using your distribution's supported package method before continuing.

tar -xzf Rabbit-Commander-Linux-1.0.0.tar.gz cd Rabbit-Commander-Linux-1.0.0 sudo bash install-linux.sh "wss://commander.zippri.com/relay" "<DEVICE_ID>" "<DEVICE_TOKEN>"
Do not put real Device Tokens into shell history on shared systems. Prefer the package's interactive wrapper when available, or clear the specific history entry after installation according to your organization's policy.

Verify / operate

SYSTEMD
sudo systemctl status rabbit-commander-agent --no-pager sudo systemctl enable rabbit-commander-agent sudo systemctl restart rabbit-commander-agent sudo journalctl -u rabbit-commander-agent -n 100 --no-pager sudo journalctl -u rabbit-commander-agent -f

Expected state: active (running) and the Rabbit Commander device shows ONLINE.

Typical VPS providers

The same Linux flow applies to ordinary Ubuntu/Debian VPS machines from providers such as Hetzner, DigitalOcean, AWS EC2, Azure VM, Google Compute Engine, Oracle Cloud, Hostinger and similar services, provided you control the OS and systemd is available.

Server hardening

Operations R3: Choose Safe Auto under Approvals for narrow automatic diagnostics. A customer-project runner needs its own reviewed broker. G2 C1-01 remains restricted; no broad root access is granted.
Dedicated OS accountUse the least privilege needed for your workload.Allowed rootsRestrict Rabbit Commander file access to required directories.Command approvalsKeep state-changing shell operations behind the approval gate.Outbound-only networkingThe agent normally does not need a new inbound firewall rule.
RUNPOD GPU

RunPod setup with persistent storage

RunPod containers can be replaced or restarted. Keep the Rabbit Commander agent and its configuration on a persistent volume (normally under /workspace) and use the supervisor pattern when systemd is not available.

GPU POD/workspacePersistent recommended

Fresh RunPod installation

Download RunPod Installer
1
Attach persistent storage

Mount your RunPod Network Volume or persistent storage at /workspace before installing.

2
Check Node.js

Run node --version. Version 18 or newer is required.

3
Run the interactive installer

The installer asks for the Device ID and Device Token without putting the token directly in the command line.

chmod +x install-rabbit-commander-runpod.sh bash install-rabbit-commander-runpod.sh
4
Verify ONLINE

Open Rabbit Commander and confirm the RunPod device becomes ONLINE.

Persistent RunPod layout

Persistent directory/workspace/rabbit-commander-agentEnvironment file/workspace/rabbit-commander-agent/agent.envSupervisor/workspace/rabbit-commander-agent/supervise-rabbit-agent.shSupervisor log/workspace/rabbit-commander-agent/agent-supervisor.log

The installer keeps the Rabbit agent in /workspace so pod recreation does not erase its configuration when the same persistent volume is reattached.

RunPod startup checks

node --version nvidia-smi ls -ld /workspace

Node.js 18+ is required. GPU access is not required by Rabbit Commander itself; it only controls and inspects the authorized pod.

Important: Do not store the Device Token in notebook cells, public Git repositories or shared shell transcripts.

Repair an existing supported RunPod installation

Download RunPod Repair Scriptchmod +x rabbit-runpod-repair.sh bash rabbit-runpod-repair.sh

The repair helper stops stale Rabbit agent/supervisor processes, restarts the persistent supervisor and prints a token-redacted health check.

After RunPod restart

  1. Confirm the persistent volume is mounted at the expected path.
  2. Start the Rabbit supervisor if your template does not launch it automatically.
  3. Check the supervisor log for a successful relay connection.
  4. Open Rabbit Commander and confirm the pod device is ONLINE before running remote actions.
macOS

macOS setup

The macOS package installs a LaunchAgent with KeepAlive.

STARTUPLaunchAgentKeepAlive
Download macOS Packagenode --version tar -xzf Rabbit-Commander-macOS-1.0.0.tar.gz cd Rabbit-Commander-macOS-1.0.0 bash install-macos.sh "wss://commander.zippri.com/relay" "<DEVICE_ID>" "<DEVICE_TOKEN>"

Verify

launchctl list | grep com.rabbit.commander.agent

The agent directory is stored under ~/Library/Application Support/RabbitCommander/Agent.

CHATGPT

Connect Rabbit Commander to ChatGPT

Rabbit Commander exposes a public remote MCP server. ChatGPT connects to the MCP endpoint; the agent remains installed on the controlled device.

MCPOAuthRemote endpoint

Connection steps

CURRENT UI
  1. In ChatGPT, enable Developer Mode/custom MCP access when your plan and workspace policy allow it.
  2. Open the app/connector creation flow from the Apps settings.
  3. Name the app Rabbit Commander.
  4. Use https://commander.zippri.com/mcp-native-v1 as the MCP endpoint.
  5. Select OAuth authentication and complete the Rabbit Commander sign-in/authorization flow.
  6. Run the tool scan and create/save the app.
  7. Enable/select Rabbit Commander for a message, then test: Using Rabbit Commander, call rabbit_list_devices.

Availability notes

Business / Enterprise / EduOpenAI documents full MCP support, including write/modify actions, in developer mode, subject to workspace permissions and admin controls.ProOpenAI documents custom MCP access with read/fetch permissions; full write MCP availability is more restricted.Remote serverChatGPT connects to remote MCP servers. Rabbit Commander's production endpoint is public HTTPS.
OpenAI official MCP guide
CLAUDE

Connect Rabbit Commander to Claude

Use Claude's custom remote connector flow. The remote connection is brokered from Anthropic's cloud.

CONNECTORRemote MCPOAuth
  1. Open Claude Customize -> Connectors.
  2. Select Add custom connector. Team/Enterprise organizations may require an Owner to register it first.
  3. Name it Rabbit Commander.
  4. Use https://commander.zippri.com/mcp-native-v1.
  5. Complete Rabbit Commander OAuth.
  6. Enable the connector for the conversation.
  7. Test: Using Rabbit Commander, call rabbit_list_devices.

Claude availability

Anthropic currently documents custom remote MCP connectors for Free, Pro, Max, Team and Enterprise plans; Free accounts are limited to one custom connector. Remote connectors must be reachable from Anthropic's cloud.

Claude official remote MCP guide
GEMINI CLI

Connect Rabbit Commander to Gemini CLI

Gemini CLI supports remote MCP servers with OAuth discovery and the /mcp auth flow.

VERIFIED BASELINE0.61.0gemini-3.5-flash-lite

Rabbit Commander bundled Windows CLI

& "C:\ProgramData\RabbitCommander\Tools\Gemini\gemini.cmd" --skip-trust --model gemini-3.5-flash-lite/mcp auth rabbit-commander/mcp listUsing Rabbit Commander, call rabbit_list_devices.

Expected baseline: rabbit-commander - Ready (21 tools).

OAuth behavior

Gemini CLI can discover OAuth metadata for remote MCP servers, open a browser for authorization, store tokens for future use and refresh them when the server provides refresh tokens.

Headless warning: interactive OAuth normally requires a local browser and localhost callback. If you run Gemini CLI only over SSH on a headless server, perform the OAuth-capable client flow from a machine that can open the browser/callback or use an approved alternative client setup.
Gemini CLI MCP documentation
OTHER AI CLIENTS

Other MCP-compatible clients

Use this only for clients that support remote HTTP MCP plus OAuth.

ENDPOINTMCPOAuth + PKCE
Preferred MCP endpointhttps://commander.zippri.com/mcp-native-v1Compatibility endpointhttps://commander.zippri.com/mcpOAuth issuerhttps://commander.zippri.comScopesmcp:read mcp:write offline_accessRelay used by device agentswss://commander.zippri.com/relay

Acceptance checklist

  1. Client connects to the MCP URL.
  2. OAuth completes successfully.
  3. Tool discovery succeeds.
  4. rabbit_list_devices returns only devices owned by the authorized Rabbit Commander account.
  5. A read action works.
  6. A write/shell action requests the expected approval before execution.
TROUBLESHOOTING

Fast diagnosis

Identify whether the failure is the local agent, the relay, OAuth, MCP tool discovery or an approval/security policy.

HEALTHCHECKNo secrets
Device OFFLINECheck the local agent/service, outbound WSS to commander.zippri.com, system time and the device token configuration. ChatGPT/Claude cannot discover toolsReconnect the custom app/connector, confirm the exact MCP endpoint and complete OAuth again if prompted. Gemini says authentication requiredRun /mcp auth rabbit-commander, then /mcp list. Read blockedCheck the device Allowed Roots and security profile. Write/command waitingOpen Rabbit Commander Approvals and approve the exact pending action fingerprint. Windows terminal keeps blinkingReinstall the current Windows EXE, close the old exited terminal session, then verify the scheduled task and device ONLINE state.

Safe support bundle

  1. Rabbit Commander version.
  2. Operating system and version.
  3. AI client and version/plan.
  4. Device name only - not the token.
  5. Approximate time of failure.
  6. Exact Rabbit tool name.
  7. Exact error text with all secrets removed.
Run built-in diagnosticsOpen support guide