Connect your own AI to your own servers.
ServerAlly can be used by the assistant you already pay for — Claude, ChatGPT, or anything that speaks the Model Context Protocol. You connect once in a browser, choose what it may do, and it can then work on your servers by asking ServerAlly rather than by being handed a password.
Connecting
Add https://serverally.firevps.net/mcp as a connector in your assistant. It will send you to ServerAlly to sign in and approve the connection. Nothing is shared until you approve it, and your assistant never sees your ServerAlly password.
Connections are listed in Settings under Connected applications, with what each one may do. Revoking a connection there stops it working immediately.
What it may do is your choice
Read-only
Looks and reports. Your assistant can list your servers and websites, read health readings and logs, and see the results of security and malware scans. It cannot change anything. If someone gets hold of this connection, the worst they can do is read.
Full access
Everything above, plus doing the work: create a website, add a database, turn on HTTPS, deploy code, run a prepared script, read and change a site's settings file. Bounded actions with known effects — there is no general-purpose command line at this level.
Full power
Adds a real command line on your servers. Catastrophic commands are still refused outright, and every call is recorded — but this is unbounded access, and the honest description is that you are trusting your assistant the way you would trust yourself.
Missions, and the approval step
A long job — an investigation, a clean-up, a migration — can be given a name and a step budget. From then on every command your assistant runs is written into that mission's record automatically. You watch it fill in on the Missions page while it happens.
Inside a mission, a command ServerAlly rates risky does not run. Your assistant is handed a one-time link, you read the exact command on ServerAlly's own page, and only then does it run — once. Approving in the chat does nothing at all. That is on purpose: consent that travels back through the assistant is consent the assistant could have reworded.
A mission your assistant finishes is recorded as not independently checked, because in this lane there is no AI of ours to verify it. We would rather say that than show you a tick we did not earn.
What your assistant is not given
Your server passwords and keys are never sent to it. It asks ServerAlly to act, and ServerAlly holds the credentials.
Files and logs are secret-masked before they reach it. The one exception is a site's settings file, which exists to hold credentials — reading that needs Full access, so a Read-only connection cannot be talked into it.
Everything that comes back from one of your servers is labelled as information rather than instructions. Servers get compromised, and a file on a compromised server can contain text written to fool an assistant into running something. Marking the boundary is how that gets treated as data.
Without an AI at all
None of this is required. Ready-made jobs, the script library, monitoring, backups, deployments and the security scanner all work with no AI key and no assistant connected. The AI is a way in, not the product.
Questions: support@firevps.net · How we handle root access · Privacy