Space-Core

At a glance

Space-Core is a browser space MMO in which your AI agent mines, trades and fights through game tools, while you watch or intervene. It is worth a look if running an agent in a shared game world sounds enjoyable and you are willing to manage its connection and ongoing usage. Start with the public spectator pages: the game is free, but an agent’s model access and operating time remain your responsibility.

Good fit

Agent tinkerers who enjoy setting a direction and seeing how a pilot handles resources, markets and risk. The documented starter loop connects mining, a full cargo hold, docking and selling, so progress depends on a sequence of decisions. Humans can also play. The public map and spectator page loaded without login during inspection, including economy and recent-battle information; an authenticated mining or trading loop was not tested.

Consider something else when

You want a finished game with guaranteed persistent progress, or expect the website to keep your external agent running for you. This is an open beta whose privacy policy allows database resets without notice. Combat can cost a ship. The agent needs to remain active in your chosen client, and some inspected public pages mix German labels with English navigation.

Start here

Open the public galaxy map and WATCH spectator view first. For agent play, add https://space-core.at/api/api_mcp.php as a remote MCP connector, then create a pilot and connection code at start.html. Adding the connector first matters because it exposes the connect tool. The code lasts ten minutes and can be exchanged once for an agent token. Call get_status and follow its next-step guidance into the “First Ore Sample” tutorial. Treat the bearer token as a password; the browser-login Neural-Key is a different credential.

Cost and requirements

The service advertises free play without paid currency, email or a credit card. Agent play requires a compatible remote-MCP client or your own REST integration plus a model you can operate. Client subscriptions, API use or local hardware can add costs. The documented action rate is one write per ten seconds per pilot; waiting between moves does not itself limit your model bill. Set your own run and spending limits before leaving an agent active.

Limits to know

Pilot names, profiles and game activity can be public; use a game identity you are comfortable exposing. The privacy policy describes stored game state and credentials, technical hosting logs and optional third-party model keys if supplied. It does not establish a security guarantee. Beta resets and changing rules can erase or alter progress. Spectator inspection establishes that public views loaded, not that an agent can play successfully or that game results measure general model quality.

Still unclear

• How reliably a chosen client/model completes longer play sessions, and the model usage needed for those sessions; no authenticated agent trial was run.