No root password in the chat
Your AI agent on the server, without the keys in hand
SSH MCP connects Claude, ChatGPT or Cursor to your servers. The agent runs commands, reads and edits files and manages servers, while the root password and SSH keys never leave the connector. Risky commands do not run until you approve them.
Giving AI server access today usually means giving it the password
AI tools have become good enough to maintain a server on their own. What is missing is a safe way for them to reach it.
- The root password ends up in the chat. And conversations are stored, shared and kept longer than you think.
- The key is on your laptop, the agent is in the cloud. Scheduled tasks and remote agents cannot reach your computer.
- There are no brakes. An agent with full access can wipe a whole site with one wrong command.
- Every server is set up separately. And again for every AI tool, with new copies of the keys.
What the agent can do through SSH MCP
The agent gets a set of tools for working with servers. You ask in plain language, and the connector opens the connection, runs the command and returns the result.
Any command
The agent runs any shell command on the chosen server and gets the output, errors and exit code.
Risky commands blocked
Deletions, overwriting system files and similar commands are stopped and wait for your explicit approval.
No passwords in the chat
Access keys live on the connector. The AI tool only sees the server name, never a password or key.
Server list
The agent sees a list of your servers with descriptions and picks the right one by name, with no IPs or ports in the chat.
Adding servers
A new server is added with its address, user and port, and is instantly available to all your AI tools.
Files both ways
Reading, writing, uploading and downloading files over SFTP, even on servers without shell access.
The agent gets the result, the key stays with you
SSH MCP sits between your AI tool and your servers. The agent never talks to a server directly.
You add your servers
Each server gets a short name, such as prod or server2, along with its address, user and port. We set up the access key, or you use your own.
You connect your AI tool
You get the connector address and short instructions for Claude, ChatGPT, Cursor or whatever tool you use. One connector covers them all.
The agent works by server name
The agent sends only the server name and the command. The connector opens the SSH connection with its own key, runs the command and returns the output.
Risky commands wait for you
When a command could cause damage that is hard to undo, the connector does not run it and asks for approval instead. Approval applies only to that exact command and expires after a few minutes.
Everything is logged
Every command is recorded: when, on which server and with what result. You switch access off in one step, without changing passwords on your servers.
One monthly price, no per-command fees
You pay monthly and get the connector, risky command blocking and server management. No per-command, per-user or per-AI-tool fees.
- Works with your existing servers. A VPS or dedicated server with any provider, anything with SSH.
- Works from the cloud too. Scheduled tasks and remote agents reach your servers even when your computer is off.
- One connector for every tool. Use the same servers from Claude, ChatGPT and Cursor.
Everything included. No per-command fees and no long-term contract.
- Run any command over SSH
- Risky command blocking and approval
- Add and list servers
- Files over SFTP
- ChatGPT, Claude, Cursor and other MCP tools
- Setup and support in English and Serbian
For everyone who runs servers and works with AI
- Agencies maintaining dozens of client sites across several servers
- Developers working with Claude Code, Cursor or Codex
- System administrators handing routine checks to AI
- Companies with their own VPS and no full-time admin
- Teams with scheduled AI tasks that need to run at night
- Anyone with their own servers who does not want to move to new hosting
SSH MCP, the most common questions
How does the agent reach the server without a password?
The connector holds the access key and opens the SSH connection on the agent's behalf. The agent sends only the server name and the command and gets the command output back. The password and key never pass through the conversation.
Which commands count as unsafe?
Deleting files and folders, overwriting sensitive files, stopping services and servers, changing users and permissions, and other commands that can cause damage that is hard to undo. Those wait for your explicit approval, while everything else runs right away.
What does approval look like?
The agent tells you in the chat which command it wants to run and why. Once you confirm, it gets a one-time approval for that exact command only, which expires after a few minutes. For the next risky command it asks again.
Does it work with servers at other providers?
Yes. SSH MCP works with any server that has SSH access, wherever it is. You do not have to move your server.
What is the difference between SSH MCP and AI Ready Hosting?
SSH MCP is for your existing servers, wherever they are. AI Ready Hosting is our hosting with MCP access already built in. If you have your own servers, take SSH MCP. If you also need hosting, take AI Ready Hosting.
What do I get for €49 a month?
The SSH MCP connector, command execution, risky command blocking and approval, adding and listing servers, file work over SFTP, setup and support. No per-command fees and no long-term contract.
Let's see what is worth automating in your business
On the first call we go through your processes and tell you honestly where AI makes sense and where it does not. No obligation and no proposal you did not ask for.