For developers

Let your AI agent work on real servers. Keep the undo.

Cloudnan puts an agent on your server and an MCP endpoint in front of it. Claude Code or Codex get real tools for deploys, configs, databases and logs. You get a preview, an approval step for risky actions, an audit log and a one-click undo.

Bring your own server Connect your cloud
your server
$ curl -sSL https://cloudnan.com/downloads/install.sh \
    | sudo bash -s -- --token <your-token>
✓ Agent installed, connected over mTLS
your agent's MCP config
{
  "mcpServers": {
    "cloudnan": {
      "url": "https://cloudnan.com/api/mcp",
      "headers": { "Authorization": "Bearer cnmcp_<your-token>" }
    }
  }
}

From bare server to agent-ready in four steps.

  1. Connect a server

    Run one install command on any Linux server, or let Cloudnan create one in your DigitalOcean, Hetzner, Vultr or Linode account.

  2. Create an MCP token

    Generate a token in the dashboard. It is revocable and never carries an SSH key.

  3. Add Cloudnan to your agent

    Claude Code, Codex, Hermes and OpenClaw discover Cloudnan's server tools over MCP.

  4. Work with an undo

    Every action is previewed, labelled, approved when risky, audited, and undoable in one click.

Why not just give the agent SSH?

A root shell lets the agent do anything, and leaves you reading shell history to figure out what happened. Through Cloudnan the agent works with typed tools Cloudnan can preview and reverse:

  • Changes labelled Undoable, Partial or Not undoable before they run
  • Risky actions held for a human approval
  • Automatic rollback when a health check fails after a change
  • Every call in the audit log under the agent's name
  • Access is a revocable token, not a root key copied around
Read about Cloudnan MCP
79+ TOOLS

Every server operation, behind one chat.

The AI Assistant ships with a curated library of production-grade tools — each parameter-validated, risk-tiered, and audit-logged. Ask in English, get the right tool with the right scope every time.

Read-only diagnostics

18 tools

  • get_metrics, get_load_history, get_io_stats
  • search_journalctl, get_kernel_messages
  • list_processes, list_services, list_listening_ports
  • find_large_files, find_recently_modified

Deploy from GitHub

5 tools

  • inspect_github_repo — read package.json, detect framework
  • check_runtime_installed — verify Node / PHP / Python / Go
  • deploy_from_github — clone, build, run, nginx + SSL
  • redeploy_app — pull HEAD, rebuild, zero-downtime swap
  • destroy_app_deployment — clean teardown

Databases

9 tools

  • list_postgres_dbs, get_postgres_connections, vacuum_postgres_db
  • get_mysql_processlist, get_mysql_slow_queries
  • redis_info, redis_dbsize

Docker

4 tools

  • list_containers, inspect_container
  • get_container_logs, restart_container

Security audit & scan

8 tools

  • list_security_findings, security_scan_folder, security_scan_container
  • list_sudoers, find_world_writable, find_setuid_binaries
  • get_ssh_login_history, list_authorized_keys

Network & firewall

5 tools

  • list_firewall_rules, network_stats
  • get_failed_logins, block_ip
  • list_listening_ports

DNS via Cloudflare

3 tools

  • check_cloudflare_integration — auto-detect connected account
  • lookup_dns_record — read existing A / CNAME / TXT
  • create_dns_record — add records straight from chat

Apps & frameworks

5 tools

  • list_pm2_processes, get_pm2_logs
  • wp_check_updates — WordPress core + plugins
  • php_info_summary, check_php_errors

Maintenance

6 tools

  • list_pending_updates, apply_security_updates
  • list_cron_jobs, list_systemd_timers
  • find_large_files, apply_all_upgrades

Process control

6 tools

  • service_control — start / stop / restart
  • kill_process, restart_service
  • install_package, add_firewall_rule, rotate_logs

Websites & nginx

4 tools

  • get_site_status, check_ssl_cert
  • test_nginx_config, reload_nginx

Mail + extras

6 tools

  • check_smtp_blacklist, get_mail_queue
  • list_wg_peers, list_backup_archives
  • list_sites, list_databases (control plane)

New tools ship monthly. Every tool is open-source — audit the registry on GitHub.

100+ SAFETY CHECKS

Guardrails so you can hand it the keys.

Letting AI run commands on production servers is only safe if every dangerous operation gates on a human and every move is auditable. We built Cloudnan that way from day one.

Tier · Auto69

Read-only & safe

Reads, lookups, listings, diagnostics. Run instantly without prompting — same blast radius as a SELECT.

get_metrics · list_processes · search_journalctl

Tier · Approve16

Mutating — your call

Anything that writes to the system. AI prepares the call, you click Approve / Deny. No quiet mutations, ever.

restart_service · install_package · deploy_from_github · block_ip

Tier · Danger0

PIN-gated

Destructive operations (drop, destroy, force-delete) require a second factor on top of approval — or are flagged as not-yet-supported.

reserved for: drop_database · destroy_vm · wipe_volume

mTLS-secured agent

Every agent ↔ control-plane message is mutual-TLS encrypted with per-agent certs. Not terminated at a third-party edge.

Per-tool parameter validation

Every tool ships with a typed schema (string / int / enum / bool). Invalid params reject before reaching the agent.

Command blocklist

The agent vetoes destructive shell patterns even when the AI generates them — rm -rf /, mkfs, dd if=/dev/zero, etc.

Risk-tier gating

Auto / Approve / Danger. The HTTP layer enforces the gate; only Auto runs without a human in the loop.

Org isolation

Every query is scoped by the user's active org. The AI cannot read or touch a server outside the membership.

Audit log on every invocation

Tool name, params, status, output — persisted with the conversation ID and your user ID for post-incident review.

Rate limit + credit gate

Each AI message debits one AI credit, capped per plan. Loops can't run away with your wallet.

Open-source agent

The binary that runs on your servers is on GitHub. Every command path, every byte sent upstream — readable and forkable.

Reversible by default

Mutations log enough state to roll back: deploy logs the previous commit, install_package logs version pinning.

Zero-Trust Foundation

Built with security at the architecture level.

No Open Ports on Your Server

Agents connect OUT only. Zero inbound attack surface on your servers. Passes security audits and firewall reviews.

mTLS on Every Plan

Mutual TLS for agent ↔ control-plane on every subscription tier. Per-agent certificates, end-to-end, no third-party proxy in the path.

Open-Source Agent

The binary that runs on your servers is on GitHub. Audit every cryptographic operation, every byte sent upstream — verify, fork, or build your own.

Configuration Drift Detection

Cloudnan tracks every package, service, and container it installs. Anything that changes outside the panel triggers an alert with one-click reconcile.

We Never Touch Your Code or Data

Cloudnan manages server operations. We never read your application code, database contents, or business data.

Role-Based Access Control

Per-org granular permissions. Assign exactly the access each person needs — no more, no less.

Full Audit Logs

Every panel action timestamped, attributed, and exportable. Built for SOC 2, ISO 27001, and similar review processes.

Multi-Factor Authentication

TOTP, email OTP, and passkey (WebAuthn) supported for dashboard access. Sign-in with Google or GitHub for OAuth-first workflows.

Encrypted Backups

Backup destinations (S3, MinIO, custom) encrypt with your own key before upload. Cloudnan never stores plaintext copies — your data, your encryption.

Pricing

Pay for the software. Not for servers.

You pay Cloudnan for the software. Your servers are billed by your own cloud provider, at their price.

Free

For personal use and exploring.

$0/mo
2Servers
1Members
1Orgs

Starter

For hobbyists and freelancers

$9/mo
5Servers
1Members
1Orgs
Popular

Pro

For small teams and agencies

$29/mo
15Servers
5Members
3Orgs

Business

For growing companies

$79/mo
100Servers
20Members
15Orgs
Compare every plan feature

Put your servers on autopilot.

Connect your cloud account or bring any Linux server. Every change gets a preview and an undo, including the ones your AI agent makes.

Connect your cloud Bring your own server
  • 7-day free trial
  • No credit card
  • Your provider bills your servers
  • Cancel anytime

Running many client sites? Talk to us