@carverjs/multiplayer
v0.0.4
Published
Serverless P2P multiplayer for CarverJS games — WebRTC mesh, lobbies, host authority, and state sync over MQTT or Firebase signaling.
Downloads
474
Readme
@carverjs/multiplayer
Serverless peer-to-peer multiplayer for CarverJS games. A WebRTC data-channel mesh with pluggable signaling (MQTT or Firebase), lobbies, host authority, and three sync engines — no game server required.
Beta: CarverJS is under active development. APIs may change between minor versions until 1.0.
Install
npm install @carverjs/multiplayer
# optional — Firebase RTDB signaling (MQTT is the zero-config default)
npm install firebasePeer dependencies: @carverjs/core, @react-three/fiber, react, react-dom. firebase is an optional peer, needed only when you choose the Firebase strategy.
How it works
Signaling (MQTT or Firebase) is used only to introduce peers and relay SDP/ICE. Once the WebRTC connection is established, all game data flows directly peer-to-peer over data channels — the signaling backend never sees gameplay traffic. One peer acts as the authoritative host; host migrates automatically if it leaves.
Quick start
Wrap your game in a provider, join a room, and exchange typed events:
import {
MultiplayerProvider, MultiplayerBridge,
useRoom, usePlayers, useNetworkEvents,
} from "@carverjs/multiplayer";
import { Game, World } from "@carverjs/core/components";
function App() {
return (
// Zero-config: free public MQTT brokers handle signaling
<MultiplayerProvider appId="my-game">
<Game mode="2d">
<MultiplayerBridge>
<World>
<Lobby />
</World>
</MultiplayerBridge>
</Game>
</MultiplayerProvider>
);
}
function Lobby() {
const room = useRoom("room-code-1234", { displayName: "Ada" });
const { players, self } = usePlayers();
const { broadcast, onEvent } = useNetworkEvents();
// room.connectionState, room.isHost, room.selfId, room.leave(), ...
return <span>{players.length} players · {room.isHost ? "host" : "client"}</span>;
}MultiplayerBridge connects the engine's render loop to the network layer; place it inside <Game> and around the scene that uses sync hooks.
Signaling strategies
// Free, zero-config (default): public MQTT brokers
<MultiplayerProvider appId="my-game">
// Firebase Realtime Database (bring your own project)
<MultiplayerProvider
appId="my-game"
strategy={{ type: "firebase", databaseURL: "https://your-project.firebaseio.com" }}
>STUN / TURN
Defaults to public STUN. Add a TURN relay so peers behind restrictive NATs or firewalls can still connect:
<MultiplayerProvider
appId="my-game"
iceServers={[
{ urls: "stun:stun.cloudflare.com:3478" },
{ urls: "turn:turn.cloudflare.com:3478", username: "...", credential: "..." },
]}
>TURN is only used when a direct connection fails. For same-network testing, STUN alone is enough.
Connection reliability
The WebRTC mesh self-heals while peers are connecting — no configuration required. Each link has a single deterministic initiator (the peer with the lower id); if a pair hasn't reached connected shortly after the first offer, the initiator automatically re-sends it with an ICE restart, a few times over roughly 20 seconds, until the link comes up. This transparently recovers the transient failures of serverless signaling:
- an offer, answer, or ICE candidate dropped in transit;
- ICE candidates that arrive before the peer is ready (buffered, not discarded);
- slow STUN candidate gathering on a cold network.
Every pair in the mesh establishes independently and recovers on its own, so a hiccup on one link no longer leaves two players unable to see each other — the rest of the room is unaffected and the stalled pair re-handshakes itself. This makes STUN-only deployments reliable in practice. A TURN relay is still required only for pairs that genuinely can't traverse each other's NAT (e.g. two symmetric/CGNAT endpoints) — STUN cannot relay those no matter how many times the handshake retries.
Sync modes
useMultiplayer({ mode }) selects how world state is replicated:
| Mode | Use it for | How |
| --- | --- | --- |
| events | turn-based, sandbox, chat, infrequent state changes | typed messages over a reliable, ordered channel |
| snapshot | real-time movement | host broadcasts delta-compressed snapshots; clients interpolate |
| prediction | fast-paced action | client-side prediction with server reconciliation and rollback |
import { useMultiplayer } from "@carverjs/multiplayer";
function Scene() {
useMultiplayer({ mode: "snapshot", tickRate: 60 });
// ... actors marked networked are replicated automatically
}Hooks
| Hook | Purpose |
| --- | --- |
| useRoom(roomId, opts) | Join / leave a room; exposes connection state, host, and self id. |
| useLobby() | Browse advertised rooms. |
| usePlayers() | Live player list plus self. |
| useHost() | Host-only room controls — room state, lock, kick, host transfer. |
| useMultiplayer({ mode }) | Drive the sync engine for a scene. |
| useNetworkEvents() | Typed broadcast / sendEvent / onEvent messaging. |
| useNetworkState() | Networked spawn / despawn helpers. |
Host authority & migration
Exactly one peer is the authoritative host. If it disconnects, the engine migrates host to another peer automatically, and host election is deterministic and consistent across all peers.
To pin a specific peer as host — for example the room creator that owns the world — advertise a host priority in player metadata. The lowest priority wins; peers that advertise none rank last (preserving the default lowest-peer-id election among them):
useRoom(roomId, {
displayName: name,
playerMetadata: { hostPriority: isCreator ? 0 : 1 },
});Advanced exports
For lower-level control, the package also exports:
MqttStrategy,FirebaseStrategy— construct or inject a signaling strategy directly.NetworkSimulator— inject artificial latency and packet loss during development.InterestManager— area-of-interest filtering for large worlds.InputBuffer,computeJustPressed— input history and edge detection for prediction.DebugOverlay— on-screen network stats.
Links
- Documentation: docs.carverjs.dev
- Community: Discord
- Issues: github.com/MoneyTales/carverjs/issues
License
MIT — MoneyTales EduTech Private Limited
