← All articles
Article

A new way to connect ChatGPT and Claude to your own server

SentinelX MCP gives your assistant a real, permissioned door into your machines — not a copy-paste loop. Here is what it does, how it works, and how to wire it up.

You ask Claude to debug a failing service. It writes a beautiful diagnosis of a log file it has never seen. So you paste the log. Then the config. Then the output of systemctl status. Twenty minutes later you are a very slow terminal with extra steps.

The model was never the bottleneck. The missing piece was a way for it to reach the machine safely.

What SentinelX does

SentinelX is an MCP server that connects your assistant to real hosts — Linux, macOS, Windows. Once a machine is enrolled, you can just ask:

"Why is nginx throwing 502s on my-vps?" · "Show disk usage across all my servers."

The assistant reads the actual logs, runs the actual commands, and answers from what is really there. No pasting.

How it works

Three pieces, and the important one is the last:

ChatGPT / Claude
  ↓  MCP over HTTPS
SentinelX hub
  ↓  outbound WebSocket
agent on your host

The agent dials out to the hub, so you open no inbound ports and need no public IP. It runs as an unprivileged sentinelx user — never root.

And the part that matters: the config is deny-by-default. The assistant can only run the commands, manage the services, and edit the files you list in /etc/sentinelx/config.yaml. The allowlist — not the model's good behaviour — is the trust boundary. Every call is logged.

Connect it in three steps

1. Install the agent on the machine you want to reach:

curl -fsSL https://get.sentinelx.app | bash

Same line on Linux and macOS. Windows uses a PowerShell installer. The script prints an enrollment URL: open it, sign in, paste the token back.

2. Point your assistant at the hub — once per account, not per host:

3. Ask it something.

"List my SentinelX servers."

What it costs

One host is free, for as long as you want it. Paid plans start at $5/month when you need several machines live at the same time. The agent itself is open source (Apache-2.0), so you can read exactly what runs on your box.

Plans, docs and downloads live at sentinelx.app.

Claude using SentinelX to report a server's RAM and disk usage, noting that the free command was not in the allowlist.
SentinelX in Claude: the allowlist at work — free wasn't permitted, so the agent used df -h and /proc/meminfo instead.

The SentinelX agent is open source (Apache-2.0). Connect it to your own Linux servers in about a minute:

$ curl -fsSL https://get.sentinelx.app | bash