You want to know whether the render on your studio Mac has finished. The usual routine is to find the machine, open a remote session, remember the right command, and read the output yourself.
With noBGP connected to Claude, you ask the question instead. Claude finds the machine, runs the command, reads the output, and answers in the conversation.
noBGP gives your AI assistant a way to run commands on machines you have connected, without SSH keys, VPNs, or open ports. This guide covers what you need, how a request becomes a command, and how to use it for everyday work across one machine or many.
What you need
There are two parts: your AI assistant, and the machines you want it to reach.
Connect your assistant. Add noBGP to Claude through MCP and sign in to your noBGP account. Claude setup, Claude Code setup, and ChatGPT setup each take a few minutes.
Connect your machines. Install the noBGP agent on each machine. On macOS or Linux:
curl -fsSL https://downloads.nobgp.com/agent/install.sh | sudo sh
On Windows, in PowerShell:
powershell -NoProfile -Command "irm https://downloads.nobgp.com/agent/install.ps1 | iex"
The installer opens your browser to sign in, registers the machine, and runs the agent as a system service. There are no keys or tokens to copy. The agent supports recent versions of macOS, Windows, and most Linux distributions, including Raspberry Pi. See the full list in the agent installation guide.
Once registered, the machine appears in your noBGP account and is visible to your assistant.
Ask in plain language
You do not need to name a tool or write the command yourself. Describe what you want to know or do:
Is the render on my studio Mac finished?
Behind that answer, Claude makes a small number of tool calls through noBGP MCP:
- Find the machine. The
network_directorytool lists the networks, nodes, and services you can access, including which machines are online. - Run the command. The
commandtool runs a shell command on that node, such as listing the newest files in the renders folder. - Read the output. The command's output comes back to Claude, which interprets it and replies.
A useful answer reads like this: the newest file finished at 09:41 and is 2.1 GB. Claude can then offer the next step, such as copying the file to another machine.
Depending on your client's settings, you may be asked to approve each tool call before it runs. Reviewing the proposed command is a good habit, especially for anything that changes the machine.
Check on long-running jobs
Renders, builds, backups, and data transfers often run for longer than you want to wait at your desk. Asking about them is one of the simplest ways to use remote commands.
Try requests like:
- "Has last night's backup on the office NAS completed? Show me the end of its log."
- "Is the build still running on the workstation? How long has it been going?"
- "How far through the video export is the studio Mac?"
Ask for evidence rather than a yes or no. File timestamps, process lists, exit codes, and the last lines of a log let you judge the answer for yourself. If Claude cannot confirm something, a good response says so and shows what it checked.
Work across several machines
The same approach works when a question spans your whole setup.
See what is online. Ask "Which of my machines are online right now?" and Claude can filter the directory by online status, then summarize it by name and platform.
Run one check everywhere. Ask "How much disk space is left on each of my Linux machines?" For a group of nodes, the command_subscribe tool dispatches a command across the group and collects each machine's report, so you get one summary instead of a series of separate sessions.
Update software carefully. Plain-language fleet management is convenient, which is a reason to be deliberate. Ask Claude to try an update on one machine, confirm it worked, and then continue with the rest.
Be specific about which machines a request should touch. "My Raspberry Pis in the lab" is clearer than "everything".
Control what can run
Running commands is powerful, so noBGP checks two independent gates before any command executes. Both must allow it.
Organization roles. Your role in the noBGP organization decides whether you can request elevated execution. Only Owner and Admin roles can run commands as root or Administrator. Members cannot.
The machine's own policy. Each node has settings its owner controls:
allow-toolssets which capabilities the node serves. Removingcommanddisables remote execution and terminal access on that machine.allow-adminsets whether work may run as root or Administrator. When it is off, commands run as an unprivileged account.allow-rootslimits which filesystem paths the file tools can reach.
When a request is refused, it stays refused. noBGP does not quietly downgrade a command to a different identity. Read the node access control guide.
Prefer a terminal?
Sometimes you want to type commands yourself. noBGP can also provide browser-based shell access to a node without SSH. Read about interactive terminals.
Good habits for remote commands
- Name the machine. Clear node names make requests unambiguous.
- Start with read-only questions. Status, logs, and disk usage are a safe way to get comfortable.
- Review changes before they run. Ask Claude to show the command first when it restarts services, deletes files, or installs software.
- Keep elevated access where it is needed. Leave
allow-adminoff on machines that never need it.
Frequently asked questions
Do I need SSH, a VPN, or open ports?
No. The noBGP agent connects the machine to your noBGP network, and your assistant reaches it through noBGP MCP.
Which AI assistants can I use?
noBGP works with Claude, Claude Code, ChatGPT, and other MCP-compatible assistants. Setup guides for each are in the noBGP documentation.
Can Claude run commands as root?
Only if your organization role is Owner or Admin and the machine's allow-admin setting permits it. Otherwise the command runs as the configured user account, or is refused.
What if the machine is offline?
Claude cannot run commands on a machine that is powered off or disconnected. It can tell you the machine is offline, and the rest of your connected machines remain available.
Start with one machine
Install the agent on a machine you use every day, connect noBGP to your assistant, and ask a question you would normally log in to answer.
Set up noBGP with Claude to get started. When something breaks, see how to troubleshoot a remote machine with Claude.
