Server management has always been tethered to a desk. You SSH into a box, remember (or look up) the right flags, paste the command, and hope you got the syntax right.
And when something goes wrong at 2 AM? You're either dragging yourself to a laptop or waiting until morning and hoping the alert resolves itself. Neither is a great option when a client's production environment is involved.
MCP-powered SSH changes that. With MCP Express, you can manage your servers from anywhere, on any device — ask Claude to check your production environment from your phone, pull deployment logs from a coffee shop, or triage an alert without ever opening a terminal.
MCP Express is a hosted MCP gateway. No code to write, no Node.js server to spin up, no infrastructure to maintain. You connect your server, define what the AI is allowed to run, and you're done in under five minutes.
How MCP Express Handles SSH
MCP Express acts as a hosted MCP gateway. Instead of spinning up your own MCP server infrastructure, you connect to MCP Express, configure your SSH credentials and allowed scripts, and expose that capability to any compatible AI — including Claude, ChatGPT, and others.
Here is the basic flow:
You (natural language)
↓
AI Assistant (Claude, etc.)
↓
MCP Express Gateway
↓
SSH Connection to Your Server
↓
Command Output → Back to AI → Back to You
The key advantages of MCP Express over self-hosted MCP solutions:
- No server to deploy or maintain — MCP Express handles the infrastructure, so there's no server setup, no security patching, and no uptime monitoring on your end
- No local setup required — no Node.js MCP server to run, no ports to open on your machine
- Script allowlisting — you define exactly which commands the AI is permitted to run
- Multi-server support — manage multiple remote servers from a single chat interface, without switching tabs or tools
- Works with any MCP-compatible AI — Claude, Github Copilot, and others
Prerequisites
Before you begin, ensure you have:
- Claude — Claude.ai (web) or Claude Desktop
- MCP Express account — Free tier, no credit card required
- SSH access to your server — IP/hostname, port, username, and either an SSH private key or password
Estimated time: 5 minutes from signup to first query.
Get Started in 5 Minutes
Step 1: Create an MCP Express account
Sign up at mcp-express.com and log in to the dashboard.
Step 2: Add your SSH server
In the dashboard, create a new server.
Give it a name and description.
Go to Tools > Add New Tools.
Select SSH Command Executor. You'll need to enter the credentials to connect to your SSH server. You can either enter them manually:
- Your server's IP address or hostname
- SSH port (default: 22)
- Username
- SSH private key or password
Or paste your SSH URL to auto-fill all settings.
Credentials are stored using AWS KMS encryption — the same standard used in banking infrastructure. Rotate your credentials, and nothing needs to be redeployed.
This is the critical step. Define what the AI is allowed to run. Every command is pre-approved — the AI can only execute what you explicitly allow, giving you full control over what happens on your server.
Start conservatively with read-only monitoring scripts:
# Example: read-only monitoring scripts
df -h
free -m
uptime
systemctl status nginx
systemctl status mysql
docker ps
cat /var/log/app/error.log | tail -100
If you need the AI to execute arbitrary commands based on your prompt rather than a fixed list, you can add {{query}} to the script field instead.
Use {{query}} with caution
Adding {{query}} removes the allowlist constraint entirely — the AI will execute commands based directly on your prompts, with no restrictions. This is an advanced option suitable only for trusted, controlled environments. For most setups, a fixed allowlist is the safer choice.
Step 4: Connect Claude to MCP Express
In Claude, go to Customize → Connectors and click Add new Connector. Give it a name and set the URL to:
https://api.mcp-express.com/gateway/mcp
Claude will redirect you to MCP Express to complete the OAuth flow. Log in, grant access, and select the MCP server you just configured.
For a detailed walkthrough, see our guide on connecting MCP servers to AI via OAuth.
Once connected, you can manage your servers from any device where you're signed into Claude — laptop, phone, or tablet. No terminal app needed.
Step 5: Start managing your server with natural language
Open a conversation with Claude and try:
- "Check the system health of my server."
- "Are there any errors in the application log from the past hour?"
- "What's the current disk usage?"
Claude uses the MCP Express SSH connection to fetch real-time answers from your server.
What You Can Ask Claude to Do
Claude is particularly strong for MCP-powered SSH tasks because it doesn't just execute commands — it interprets the output and tells you what to do about it.
System diagnostics
"Check CPU, memory, and disk on the web server and tell me if anything looks concerning."
Claude runs top, free, and df, interprets the numbers, and flags anything unusual — rather than dumping raw output back at you.
Log analysis
"Look at the last 100 lines of the application error log and summarize any patterns."
Claude reads the logs and surfaces actionable insights, not just text.
Service management
"Restart the app service and confirm it came back up cleanly."
Claude sequences systemctl restart, checks status, and verifies health — all in one turn.
Deployment assistance
"Pull the latest code from git and restart the app server."
Claude handles multi-step operations that would otherwise require several manual commands.
Exploratory queries
"What processes are consuming the most memory right now?"
Claude translates intent to ps aux --sort=-%mem | head -20 without you needing to remember the exact syntax.
Claude also operates with natural caution on destructive operations. If you ask something risky, it will confirm intent before proceeding — a useful safeguard in production environments.
MCP-Powered SSH vs. Traditional SSH
| Traditional SSH | MCP-Powered SSH |
|---|
| Access method | Terminal client | Natural language via AI |
| Command syntax recall | Required | Describe intent — AI handles the syntax |
| Multi-step operations | Manual, sequential | AI sequences automatically |
| Output interpretation | You parse it | AI summarizes it |
| Setup per server | SSH config, key management | Once via MCP Express |
| Works on mobile | Rarely | Yes, via any Claude interface |
Traditional SSH isn't going anywhere — power users and complex automation will always benefit from direct shell access. But for the majority of day-to-day interactions — checking status, reviewing logs, running routine scripts — MCP-powered SSH significantly reduces the friction.
Connect SSH With MCP Today
MCP Express is currently in early access with limited seats available. We're looking for alpha users who want to help shape how this gets built.
As an early adopter, you get full access to MCP Express at no cost — including all features, unlimited servers, and priority support. Join while spots are open.
Join the MCP Express Alpha program
Frequently Asked Questions
Can I SSH from my mobile device?
Yes. As long as you're signed into Claude on your phone, you can SSH into your servers from any mobile device — no terminal app required.
Can the AI run destructive commands like rm -rf?
Only if you explicitly include such commands in your Allowed Scripts. The allowlist architecture means the AI cannot run anything you haven't pre-approved. For most setups, you'd never include destructive commands in the allowed list.
What happens if the AI tries to run a command that isn't in the allowlist?
MCP Express will reject the command and return an error to the AI. Claude will inform you that the requested operation isn't permitted under the current configuration.
Is my SSH key stored securely?
Yes. MCP Express stores credentials using AWS KMS encryption — the same standard used in banking infrastructure. Your keys are encrypted at rest and never exposed in plain text.
Can I use this with Claude Code?
Yes. Claude supports MCP server connections natively. You can write code in Claude Code and deploy it to your server in the same conversation — a natural fit for code-and-deploy workflows.
