More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Stenvakt: Server Security Agent

    2

    Securely audit and harden Linux servers from your Mac without sharing SSH keys or root access with AI agents.

    Free

    Security scanned
    Stenvakt: Server Security Agent

    Stenvakt: Server Security Agent

    Example session with this skill installed

    Is my Ubuntu server OK? Check it and tell me what to fix first.

    • Read your context and instructions
    • Compiled the stenvakt

    I ran a read-only review of web-1 through Stenvakt. Nothing was changed.

    Security 95/100 · Health 90/100. Nothing urgent: no pending updates, no pending security patches, disk at 64%, memory 75% free. Three things are worth doing, and seven are nice to have.

    Worth doing

    #AreaFindingWhy it mattersFix
    1FirewallDocker publishes port 8080 around the firewallufw does not filter ports Docker opens, so 8080 is reachable from the internet even if your rules say otherwiseIn the app: Block Docker ports from the internet
    2MemoryNo swap on a 2-core serverA short memory spike can kill your database instead of slowing it downManual: add 1 to 2 GB of swap
    3DiskDocker container logs are not rotatedThey grow until the disk is fullManual: set log rotation in /etc/docker/daemon.json

    For #3, add this to /etc/docker/daemon.json, then restart Docker:

    { "log-driver": "json-file", "log-opts": { "max-size": "10m", "max-file": "3" } }
    

    Nice to have

    • SSH: allows 6 attempts per connection. 3 to 4 is plenty with keys. The app's Harden SSH covers it.
    • SSH: listens on port 22. Moving it cuts log noise, but it is not security on its own.
    • Access: everything runs as root. A named sudo user limits what a stolen key can do.
    • Monitoring: nothing watches this server. A free uptime check plus Netdata takes five minutes.

    Recovery: SSH is the only way in. A Cloudflare Tunnel gives you a second door if a firewall change ever locks you out.

    Next step

    I can propose #1, blocking Docker port 8080 from the internet. You'll see the exact steps first, then approve or reject the change in Stenvakt. Because it's a firewall change, it runs in safe mode and rolls itself back if you can't log in afterwards. Want me to propose it?

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Managing Linux servers with AI usually means handing your agent SSH keys or root access. Stenvakt keeps the keys on your Mac and gives the agent read tools and proposal tools only: it can look at everything, but nothing changes until you approve it.

    What it does

    • Security reviews: scores server health and checks pending updates, listening ports, firewall rules and SSH settings.
    • Patching: proposes system updates and updates for installed tools like Node.js, PHP, Python, Docker, nginx and OpenSSL.
    • Threat detection: reports SSH brute-force attempts, bots, hammered ports, fail2ban bans and successful logins.
    • Hardening: proposes SSH hardening, stronger SSH crypto, fail2ban and firewall rules.
    • IP blocking: proposes blocks for attacking IPs found in the activity report. The app refuses your own and private addresses.

    How it works

    1. Read first: the agent runs review_server or audit_server. Reading changes nothing.
    2. Explain: it tells you what matters now and what can wait.
    3. Preview: it shows the exact steps, including the safety steps the app adds.
    4. Propose: it sends the change with a short reason to the Stenvakt Mac app, where you approve or reject it.
    5. Safe mode: firewall and SSH changes arm a rollback timer on the server, which restores the old settings if a fresh login fails.

    Frameworks & tools

    Works with any agent that supports MCP: Claude Code, Cursor, Codex CLI, Gemini CLI, Claude Desktop. Tested on Ubuntu and Debian servers. Needs the free Stenvakt app on macOS 14 or newer.

    Why this beats prompting it yourself

    Without it, an agent either can't reach your server or gets a raw SSH shell. This skill gives it a "propose, don't push" workflow, tells it to treat server output (logs, usernames, user agents) as untrusted data, and leans on safe-mode rollback so a bad firewall or SSH change can't lock you out.

    Use cases

    • Check a new Hetzner or DigitalOcean server for common misconfigurations.
    • Block brute-force attackers found in activity_report.
    • Install pending security updates and turn on automatic security updates.
    • Harden SSH, with an automatic rollback if you can't log back in.

    Known limitations

    The agent can't add servers, install the app or see keys. It only runs from a Mac. Safe mode covers firewall and SSH steps, not package updates or reboots, so take a snapshot before updating a production server. Deeper audits (about 60 checks) and fleet-wide audits need Stenvakt Pro.

    Needs the Stenvakt app on macOS 14 or newer. Tested on Ubuntu and Debian. The agent can't add servers or see SSH keys; you add servers in the app. Safe-mode rollback covers firewall and SSH changes, not package updates or reboots, so take a snapshot before updating a production server. Deeper and fleet-wide audits need Stenvakt Pro.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 9 days ago

    • Free with an Agensi account

    Requires

    Stenvakt Mac app
    Stenvakt MCP
    macOS 14+
    Ubuntu or Debian server
    SSH access
    Patcher MCP
    macOS

    Permissions

    Network Access

    The agent never connects to a server itself. It calls the Stenvakt MCP server, and the Stenvakt Mac app opens the SSH connection with keys the agent never sees. Reading is immediate; every change waits for the user to approve it in the app.

    Listed9 days ago

    What's inside

    Frequently Asked Questions

    No. Stenvakt uses a "propose, don't push" workflow where the AI generates security changes, but execution only happens after you manually approve them via the Stenvakt Mac app. Your keys stay in the Stenvakt app on your Mac; the agent only gets read and proposal tools.

    The skill works with any AI agent that supports the Model Context Protocol (MCP), including Claude Code, Cursor, and the Gemini CLI, provided you are running macOS 14+ to host the gateway app.

    Firewall and SSH changes run in safe mode. Before the change, the server saves its current settings and arms a rollback timer. The app then opens a fresh login; if that fails, the server restores the old settings on its own.

    It’s tested on Ubuntu and Debian servers. It can review, update and harden them, and propose updates for installed tools like Node.js, PHP, Docker and nginx.

    Yes. The skill and the Stenvakt app are free. Stenvakt Pro adds deeper audits (about 60 hardening checks) and fleet-wide audits.

    New in LLM Ops