Servers is the operational heart of HostAtlas. Everything the agent discovers — services, containers, certificates, cron jobs, disks, network interfaces — attaches to a server here. Open a row and you get the full picture of one host without ever opening an SSH session.

What you see

The index at /servers opens with three view modes, and your last choice is remembered per browser:

• Table — dense list with hostname, IP, OS, agent version, health score, last-seen, cost.
• Cards — one tile per host, colour-coded by status.
• Map — a world map with status-coloured markers. Hosts without a location are listed underneath.

A KPI strip at the top shows totals, online / offline / pending counts, and the monthly cost roll-up.

Every row has:

• Status dot — green (online), yellow (pending / stale), red (offline).
• Health score (0–100) — a rolling blend of CPU, RAM, disk, and load. Click the number to open the breakdown.
• Last seen — coloured green if under 2 min, yellow up to 5 min, red beyond.
• Agent version — struck through when a newer version is available.
• FATAL badge — appears when at least one supervisor process is in FATAL state.
• MCP toggle — an inline switch that decides whether AI assistants can reach this host through the MCP gateway.

Filters at the top narrow by OS, hoster, group, tag, and free-text on hostname/IP. Every filter is reflected in the URL, so any view is shareable.

What you can do

From the index:

• Add Server — opens the install-key modal (see below).
• Compare — pick two hosts to see them side by side.
• Groups — jump to Server Groups.
• Export CSV — export the current filtered list.

From the per-row three-dot menu:

• View details
• Open the SSH Gatekeeper request flow for this host
• Open the Attack Mode console
• Open the AI footprint report
• Add tags
• Jump to Forensic (last hour)
• Start a diagnostic capture
• Toggle MCP exposure
• Delete (with type-to-confirm)

Adding a server

Adding a server is a two-step, agent-driven flow:

1. Click Add Server. HostAtlas generates a short-lived install key.
2. Paste the shown one-liner onto the target host. The agent installs itself, registers, and reports first metrics within a few seconds. The new server appears in the list as pending, then turns green.

Servers imported from provider connections (Hetzner, Contabo, etc.) show up automatically with an Imported badge before the agent is installed. Once the agent starts reporting the badge flips to Monitored, and to Full when both sources agree.

Deleting a server

Deletes are queued, not synchronous. HostAtlas has dozens of related tables plus TimescaleDB metric and log chunks to purge, which can take up to a minute for a busy server. When you confirm the delete:

1. The row is marked as deleting — dimmed, a spinner appears, and further actions are disabled.
2. A background worker removes the data. You can navigate away.
3. The row disappears once the cascade completes. A hard refresh in the meantime shows the same deleting state, so shared links stay honest.

You must type the hostname exactly to confirm — the same pattern GitHub uses for repo deletes.

Server detail

The detail page groups everything under one host into tabs. The active tab is preserved in the URL so links keep their context.

Tab	Content
Overview	Health breakdown, real-time CPU / RAM / disk / load charts (24 h), current disks, network interfaces, key services at a glance. Hostname and IP are click-to-copy.
Disks	Every mounted disk with used / free / percent, filesystem type, and mount path.
Domains	Domains served by this host (from the discovered web-server configs), with SSL and uptime status.
Services	All auto-discovered systemd services — nginx, MySQL, PostgreSQL, Redis, MongoDB, Elasticsearch, Memcached, ClickHouse, CockroachDB, RabbitMQ, Kafka, HAProxy, Varnish, Postfix, Docker, PHP-FPM, and more. Each service exposes plugin stats and a three-dot menu to Start / Stop / Restart.
Containers	Docker containers on this host — image, ports, CPU / RAM, status, quick actions.
Certificates	SSL certs found on disk with issuer, subject, and expiry.
Queues	Redis, RabbitMQ, and Kafka queue depths.
Processes	Supervisor-managed processes with PID, uptime, and state.
Vulnerabilities	CVE findings from the light per-server scanner. See Vulnerabilities.
Firewall	Live view of the host’s firewall rules.
Cron	Every cron entry the agent sees — system cron, user crontabs, cron.d, cron.hourly / daily / weekly / monthly, systemd timers. See Cron Jobs.
Config	Read-only view of the configs the agent collects (nginx.conf, php.ini, my.cnf, postgresql.conf, …).
Audit	Findings from the configuration audit, split into Performance and Security categories. Resolve items in bulk.
Logs	Live log streaming with time-range presets and per-service dropdowns. See Logs.
Events	Chronological feed of alerts, incidents, and agent events on this host.

What you can do (on the detail page)

• Run Discovery — re-run the agent’s discovery pass on demand.
• Update Agent — remotely upgrade the agent to the latest release, one click.
• Restart a service — the three-dot menu on any service row triggers a remote restart. Actions are limited to a safe allow-list of about twenty common services.
• Network diagnostics — ping and traceroute from the host itself.
• Rotate the agent token — invalidate the current token and issue a fresh one.
• Reissue install command — regenerate the install one-liner if you lost it.
• Edit — the modal has tabs for General, Monitoring, and Custom Fields. You can pin a manual latitude / longitude here, override the auto-resolved country and city, and set cost + interval + currency.
• Attack Mode — open the incident console for this host with playbook actions.
• Forensic — jump to a time-travel view of alerts, metrics, and logs for any point in the past.
• Delete — queued, as described above.

How it works

• The agent is a single static binary. Once installed it heartbeats every 15 seconds.
• The Overview page auto-refreshes status, agent version, and last-seen every 5 seconds while the browser tab is visible.
• When the agent is more than one minor version behind, the “outdated” banner shows a one-click remote update button.
• If the agent goes offline, a reactive banner appears without a page reload.
• Geolocation is resolved from the public IP the first time HostAtlas sees the host and cached for 7 days. Manual overrides are never overwritten.

Tip

Turning MCP Off for a host is the safest way to keep an AI assistant away from a sensitive server. The toggle can only be flipped from the web UI — MCP itself can never re-enable its own access.

Configuration

Nothing on this page requires code. The tunables surfaced in the UI:

• Tags — free-form key:value labels. Used for filtering, alert-rule targeting, and recipes.
• Custom fields — arbitrary metadata (owner, ticket ID, environment).
• Cost — amount + interval (hourly / daily / weekly / monthly / yearly) + currency.
• Manual location — latitude / longitude / country / city. Used by the Map view.
• MCP exposure — per-server on/off switch for AI-assistant access.

Related

• Server Groups — bucket servers for filtering and recipes.
• Network Map — visualise how servers, services, and domains connect.
• Recipes — orchestrate actions across many servers at once.