How developers use MCP to deploy and monitor servers with AI

MCP (Model Context Protocol) lets your AI chat run a command on your server, read the output, act on it and report back — without you leaving the chat window.

What developers use MCP servers for

• Reading logs — tail a log file, filter for errors, identify patterns. Faster than grepping manually because the AI groups and explains as it reads.
• Deployments — pull code, run install, restart a service, hit a health endpoint. The AI handles the sequence and reports any step that fails.
• Health checks — disk usage, memory, CPU, failed systemd units, certificate expiry — all in one message.
• Debugging — read the error, check the config, suggest a fix and apply it if you say yes.
• File operations — find large files, clean up old logs, rename batches of files.

Local vs remote MCP servers

A local MCP server runs on the same machine as your AI desktop app. Useful for Claude Code or Cursor working with local files.

A remote MCP server runs on the internet — useful when your AI chat is a web app (ChatGPT, Claude in the browser) and your server is a VPS somewhere else. Shellway is a remote relay: install the agent on your server, add the relay URL to your AI app, and the connection is live.

Setting up MCP on a VPS

# On your VPS
npx shellway pair YOUR-PAIRING-CODE
npx shellway start

# Start automatically after reboot
npx shellway service install

Add to your AI app:

https://mcp.getshellway.com/mcp

No firewall changes. The agent connects outward over HTTPS. Auth uses OAuth — your AI app signs in to your Shellway account once.

Using MCP in Claude Code

Add Shellway to .claude/settings.json:

{
  "mcpServers": {
    "shellway": {
      "type": "http",
      "url": "https://mcp.getshellway.com/mcp"
    }
  }
}

Claude Code can now call list_devices to see your paired servers, then read files and run commands on any of them.

The tools available

Shellway provides 20 MCP tools: run_command, start_process, read_file, write_file, edit_block, search_files, list_directory, list_devices, and more. See the full tool reference.

Safety

• Run the agent as a non-root user.
• Use --readonly to disable all write and command tools at once.
• The relay stores nothing about your files or command output.

Try Shellway free →