MCSManager

MCSManager authenticates with a single API key, with no OAuth2 client and no key pair. Create one in the panel under your account, then paste it into Provibr together with the panel URL.

• Host: the full panel URL including the port, for example http://panel.example.com:23333. MCSManager runs on plain HTTP on 23333 by default, and there is no port Provibr can add without guessing.
• API key: the 32-character key the panel generates. There is one key per panel user and creating a second one overwrites the first, so give Provibr a panel user of its own rather than the account you work in every day.
• TLS fingerprint (optional): only for an https:// host, to pin the certificate.

The key must belong to an administrator: creating, changing and removing instances are admin-only routes, so a key from an ordinary user can read the panel but cannot provision anything. Provibr checks that when you test the connection, not at the first order. It is a wide mandate as well: the same key opens the panel's file manager and its user management, which is the second reason to give it a user of its own.

MCSManager serves plain HTTP on port 23333 and ships no certificate of its own. If the panel sits behind a reverse proxy with TLS, fill in that proxy's fingerprint; the agent then trusts that one certificate instead of turning verification off. Read it with:

openssl s_client -connect <host>:443 </dev/null 2>/dev/null | \
  openssl x509 -noout -fingerprint -sha256

Leave the field empty for a certificate from a public authority, and also when the panel is reachable over http only. A fingerprint next to an http address is refused, because there is no certificate to pin. A panel on plain http does work, but the API key and everything the integration does then travel unencrypted.

An MCSManager instance is in essence a start command and a working directory. There is no egg, no game code and no template with variables, so Provibr does not build a server from a catalogue entry: the start command comes from the game plan you write.

A game plan carries the start command and, if the server needs it, how to stop it. Ordering is plan-driven: what runs comes from the plan, and the daemon it lands on comes from the integration's settings, or is picked for you when the panel has only one.

• Start command: required. Nothing else decides what runs, so unlike a panel with eggs there is no default to fall back on.
• Stop command: optional. Empty means ^C, MCSManager's own notation for SIGINT and the panel's default; a Minecraft server wants stop.
• Stop timeout: optional. Empty means the panel's default of 10 seconds, after which the daemon steps in hard.

MCSManager lists its daemons and reports the total memory of each, so the placement engine has a real ceiling to rank on, more than PufferPanel and GameAP offer. On a panel with one daemon, which is the ordinary install, the agent picks it itself and there is nothing to fill in.

MCSManager suspends through an expiry timestamp on the instance, and the panel enforces it itself: the daemon refuses to start an instance whose time has passed. That makes this the honest kind of suspension: someone who logs straight into the MCSManager interface cannot start a server Provibr has suspended, which is not true of PufferPanel or GameAP.

• Suspending also stops the server. The panel's own sweep runs once an hour, so without that second step a running instance would keep running for up to sixty minutes after it was suspended.
• Resuming restores the previous expiry rather than clearing it. If an expiry was already set, wiping it would quietly turn the instance into one that never runs out.

MCSManager runs an instance in one of two ways: as a plain process in a pty, or in a Docker container. Every resource limit it has belongs to the second: memory, CPU and disk. A plain process runs without a cgroup, without a memory ceiling and without a disk quota.

There is no console: MCSManager's terminal runs over socket.io with a heartbeat of its own rather than a plain websocket, and the daemon address the panel hands out is a local one an agent elsewhere cannot reach. There are no usage graphs either. CPU and memory are real measurements, but disk is hard-coded to zero in the daemon and network is absent altogether, so two of the four charts would sit at zero forever. Reinstalling is refused with a readable reason, because MCSManager only reinstalls from a preset and Provibr does not create from one. Creating, starting, stopping, restarting, killing, reading status, suspending, resuming, deleting and importing existing instances all work.