A few steps.
A watchful server.
Install on your server. Open the panel on your computer. Connect your AI client when you’re ready.
We are testing installation through our private GitHub repository. Public downloads are not open yet. Early testers receive an authorized installation command; no GitHub password or access token needs to be saved on the monitored server.
Where is your server?
Enter the address you use for SSH. The commands below update automatically. These fields stay in your browser and are not submitted to us.
Enter your server’s IPv4 address or hostname to prepare the commands.
Enter your server address above.Enter your existing SSH password if asked. Once connected, run the preview installation command supplied to you. You do not need Node.js, Rust, MySQL or a domain name.
Installation complete. Setup key: <your one-time key> Please note this setup key for initial setup. Open another terminal on your computer. Your exact tunnel command and panel address appear here.
Copy the real setup key from your server’s terminal. It is shown only for first-time setup and expires after 30 minutes. An upgrade preserves your existing account.
Enter your server address above.Run this tunnel command and keep the terminal open. A quiet terminal is normal. It carries the dashboard securely through your SSH connection.
This address means your computer. The SSH tunnel connects it to your server. Your server’s public IP is used in the command above.
The installer keeps the dashboard private on server port 4877. It does not publish the panel through your existing websites or change their certificates. If local port 14877 is busy, use 14878 in both the tunnel command and browser address.
3. Create your account once
- Paste the setup key shown by the installer.
- Choose a username, then enter your new password twice. Use at least 14 characters.
- Confirm your authenticator code and keep the recovery codes somewhere private.
After setup, sign in with your username, password and authenticator. You will not need the setup key again.
4. Start with the included rules
Monitoring starts automatically. Open the setup checklist to review the ready-made rule packs, notifications and connections. Adjust the presets when needed; you do not have to create every CPU, memory and disk rule yourself.
A critical CPU rule, for example, checks for usage above 90% during at least 80% of a three-minute window. Brief spikes and sustained pressure are handled separately.
5. Connect your AI client
In the dashboard, open Notifications & connections → MCP clients.
- Choose A remote server and enter the same server address, SSH username and port used above. The installer suggests these details when available.
- Choose Codex or Other MCP client, give the connection a name, and click Create connection.
- Click Copy configuration. Paste the complete block into your AI client’s MCP settings, then reconnect the client.
The app creates a separate private key file for each connection and fills in the exact command and path. You do not need to paste API keys into terminal commands or edit environment variables.
SSH key sign-in must already work from the computer running your AI client. An AI client cannot answer an interactive SSH password prompt. The connection wizard explains this requirement; it does not change your SSH access settings.
Already created a connection? Generate its SSH configuration here.
This optional helper is for the standard Linux installation. Get the connection ID from the configuration shown in your dashboard. This is a non-secret UUID, not an API key, setup key or password. The dashboard’s own generated configuration is the best choice for custom installation paths.
Enter your server address and an existing connection ID.No connection is created by this website. Create it in your dashboard first.
Try asking your connected agent:
- “Show a table of CPU usage over the last 24 hours.”
- “When did CPU and memory rise together?”
- “Explain the evidence for the most recent incident.”
MCP returns collected measurements and incident evidence. Your AI client interprets them. Only history collected after installation is available. The website and dashboard addresses are not HTTP MCP endpoints.
6. Choose email and optional AI
In connection settings, add and verify your email recipient, then choose report times and instant alerts. Email requires a configured notification service; the monitored machine does not need its own mail server. Hosted public notification service availability has not been announced yet.
For optional incident analysis, choose your AI provider and model, add your API key, and click Test API connection. Monitoring and rule-based alerts work without AI. Provider charges and limits are separate.
Installing on your own Mac?
With a supplied Apple Silicon preview package installed locally, open http://127.0.0.1:4877/admin. No SSH tunnel is needed. In the MCP wizard, choose This computer.
Need a hand?
Contact services@metricowl.cloud. Never send passwords, setup keys, recovery codes or API keys in a support message.