Keep Claude Code.
Give it your machines.

AI Commander is not an alternative to Claude Code, Codex, ChatGPT, Cursor, or Grok. It is the remote-machine harness those tools plug into over MCP: outbound-only shell, detached jobs that outlive the chat, and file transfer on hardware you already own.

Quick start:
  1. Install the host agent on the machine you want your AI to control. It shows a private machine code.
  2. Connect your AI: paste this prompt, then replace AIC-… with your code — Help me connect this AI client to AI Commander using https://aicommander.dev/start.md. Then check the status of my machine: AIC-… — or add the connector URL https://aicommander.dev/mcp and sign in with an email code (per-client steps).

Sibling harness pages: Cursor · Grok · Codex · ChatGPT · Roo Code · Zed · OpenCode · Qwen Code · Continue · Gemini.

Positioning in one line: your AI stays the brain; AI Commander is the hands on the box — servers, GPU rigs, NAS, laptops — with no inbound ports and no exposed SSH.

Connect in one command

Claude Code
$ claude mcp add --transport http aicommander https://aicommander.dev/mcp
Codex
$ codex mcp add aicommander --url https://aicommander.dev/mcp

Then sign in: run /mcp inside Claude Code and choose Authenticate for aicommander, or run claude mcp login aicommander from the shell. On a machine reached over SSH, use claude mcp login aicommander --no-browser and open the printed link in a browser elsewhere. The sign-in asks for your email and a one-time email code.

ChatGPT / Claude Desktop: add custom connector URL https://aicommander.dev/mcp (one-time browser sign-in). Full client matrix: Connect your AI client.

Claude Code Remote Control and AI Commander

Claude Code’s Remote Control (Claude Code docs) lets you continue a local session from the Claude mobile app or claude.ai/code: start it with claude remote-control, claude --rc or /remote-control in an open session. Claude keeps running on that one computer, which has to stay on and online. AI Commander works alongside it: Remote Control is how you steer Claude Code from your phone, and AI Commander is what that session uses to reach your other machines — servers, GPU rigs, NAS — over MCP, including detached jobs that keep running after the chat ends. Remote Control itself needs a claude.ai Pro, Max, Team or Enterprise login, not an API key. No extra AI Commander setup is needed: run claude remote-control in the project where you ran the claude mcp add command above (or add it with --scope user), and the session you drive from your phone still has AI Commander’s tools.

What you get that the chat alone cannot

Not a replacement — a multiplier

You already useAI Commander adds
Claude CodeMCP remote exec + jobs on your fleet
Grok / grok CLIgrok mcp add --transport http … → remote exec + jobs
CodexSame MCP endpoint; aliases after sign-in
ChatGPTCustom MCP connector or plain REST /api/v1/exec
Cursor / other MCP clientsnpx @aicommander/mcp stdio bridge

FAQ

Is AI Commander an alternative to Claude Code?
No. You keep Claude Code as the AI. AI Commander gives it remote shell, detached jobs, and file transfer on machines you own over MCP.
How do I connect Claude Code to AI Commander?
Run: claude mcp add --transport http aicommander https://aicommander.dev/mcp — then install the agent on the target machine and ask Claude to use the machine code or saved alias.
Can I run Claude Code on my own box?
Yes. Install the host agent on the box; it connects outbound, so no inbound ports are needed. With Claude Code installed and logged in there, ask your AI client to start claude -p on that box as a detached job (remote_job_start), then collect its transcript and exit code later (see the Remote coding agents guide). Claude Code itself can also be the client that drives your boxes over the AI Commander MCP server.
How many Claude Code Remote Control sessions can I run, and how does AI Commander fit in?
Per the Claude Code docs, outside server mode each Claude Code process supports one remote session at a time, so two interactive claude processes mean two separate remote sessions. Server mode (claude remote-control) runs several concurrent sessions from a single process, up to 32 by default (--capacity). Both modes run on the one machine where claude is running. AI Commander works alongside them: it gives the session you drive from your phone or claude.ai/code a remote shell, detached jobs and file transfer (Pro) on your other machines over MCP, and it does not replace Remote Control.
Why does claude remote-control --sandbox or --no-sandbox exit with an error?
Because both flags were removed. Per the Claude Code Remote Control docs, the --sandbox and --no-sandbox flags were removed in Claude Code v2.1.300: if you pass either one to claude remote-control, the server exits with an error instead of starting. To sandbox the sessions the server starts, turn on sandboxing in a Claude Code settings file instead. If an old script or wrapper still passes either flag, remove it. This is a Claude Code change; removing the flags does not change how AI Commander's MCP tools connect.

Also see the agent stack for Claude Code, Codex & Grok users — AI Commander plus PLAI.chat, MCPfinder and cnvs.app, each priced separately (no bundle). The three MCP servers for Claude Code with exact commands: MCP stack for Claude Code.

Running Claude Code against a VPS? See Claude Code on a VPS without SSH: outbound-only host agent, no open ports, detached jobs and logs.

Next steps

Install the agent on a machine you control, connect Claude Code / Codex / ChatGPT, then ask for a real task.

Install the agent Setup guides Remote coding agents Agent stack