Build, run and publish 3D games in the Zero engine from Claude Code, Cursor or Codex, over MCP.
Copy the AI prompt to install this server into Claude Code, Cursor, or another agent โ or use 1-click editor setup below.
We haven't yet run this listing's install command through our automated sandbox check. This isn't a red flag โ we're steadily working through the catalog.
๐ก Paste the JSON block into your client's configuration file under mcpServers, then restart the application.
@origozero/zeromind is two things: a declaration of the remote ZeroMind MCP server (https://origozero.ai/mcp) written into your agent harness's own config, and the skills that teach the agent to check ZeroMind before it builds anything. Nothing runs on your machine โ the CLI links the machine to your ZeroMind account, writes the server entry, and exits. Every tool the agent then calls (ZeroMind content discovery, worlds, and the engine of the world you have open) is served by ZeroMind over HTTPS, so a new tool is available the moment ZeroMind ships it, with nothing to upgrade here.
Each supported harness (Claude Code, Cursor, Codex CLI, Gemini CLI, OpenCode, Cline, Continue, Windsurf, Zed, openClaw, JetBrains Junie, Sourcegraph Amp, GitHub Copilot, Block Goose, Aider) gets its onboarding through that harness's own native channel โ skills, AGENTS.md, GEMINI.md, .cursor/rules/*.mdc, .clinerules, CONVENTIONS.md, .goosehints โ plus the MCP entry in the config file that harness reads. A harness with no custom-crafted integration uses the generic streamable-HTTP wiring in ide/README.md.
What the agent gets, once linked: the ZeroMind library (search, inspect, preview, install, engage), worlds (list, create, fork, launch, connect, and the reversible delete / trash / restore), and โ for the world it is connected to โ the engine itself (execute, guides, capture, the VFS, bash). tools/list on https://origozero.ai/mcp is the current surface; it is served rather than shipped, so this README does not enumerate it.
Node.js 18 or newer, on your PATH, to run zeromind link, zeromind install, zeromind upload and zeromind open. Your harness then talks to https://origozero.ai/mcp over HTTPS and spawns nothing locally.
brew install node (or download from https://nodejs.org)Verify with node --version.
The bundle carries the skills and an .mcp.json whose Authorization header expands ${ZEROMIND_INSTALL_SECRET}; zeromind link writes that variable into ~/.claude/settings.json env. Restart Claude Code afterwards so the new setting is read.
One command per harness โ it links the machine first if it isn't linked yet, writes the harness's native instruction file, and writes the /mcp entry into that harness's own MCP config:
Every command is idempotent: shared files (AGENTS.md, GEMINI.md, CONVENTIONS.md, .goosehints, copilot-instructions.md) carry a delimited <!-- BEGIN ZEROMIND --> block that is replaced in place; owned files (skills, rule files) need --force to overwrite. Two harnesses have no MCP config path to write into โ openClaw's is unconfirmed, and Aider takes servers on the command line โ so the install prints their entry for you to paste. ide/README.md has the full table, with a README per harness.
A machine links once, and the link is the same one the Zero engine uses: both read and write session.json in the OS config dir (~/.config/zero/session.json; %APPDATA%\zero\session.json on Windows; mode 0600 off Windows). Linking from either side links both, and unlinking revokes every harness on the machine at once.
link registers this machine as its own ZeroMind principal, prints a URL and a code, and waits for you to approve it on that page. The approved account is the agent's identity โ its own ZeroMind account, distinct from yours.
On a machine that already holds an approved install โ linked by an earlier run, or by the Zero engine, which writes the same file โ link prints who the machine acts as, asks for no new code, and still hands the secret to Claude Code. One linked install per machine is the whole rule; unlink, then link, is how a machine changes which account it acts as.
unlink revokes the install at ZeroMind and removes it from this machine's cache โ the install fields only, so a session the Zero engine signed in with, and the issuer it signed in against, stay where they are. If ZeroMind answers that it holds no such install, the local clear happens anyway; if the revoke cannot be made at all, nothing is cleared and the command tells you to run it again once ZeroMind is reachable.
unlink also takes ZEROMIND_INSTALL_SECRET back out of ~/.claude/settings.json when it holds the secret being revoked โ before it revokes anything, so a settings file it cannot read is reported and the unlink still completes. (That one line is then yours to delete.) The entries install <harness> wrote into other harnesses keep the revoked secret and answer 401 until you link again and re-run the install for that harness โ the command says so when it runs.
The credential a harness presents is that install's secret, sent as Authorization: Bearer ins_sec_โฆ. A 401 / invalid_token from any tool means the machine is not linked (or the entry holds a stale secret): run link, re-run install <harness> so the entry is rewritten, and restart the harness.
Copies a file โ or a folder, its relative layout preserved โ from this machine into the world's engine VFS, under /source unless --to says otherwise. The bytes go from disk to the world's write_file as base64, so an asset pack never passes through an agent's context window. Ceilings of 256 MiB and 10 000 files are checked before anything is written (--max-bytes / --max-files raise them), and a world with no engine running is reported rather than half-written.
Every tool an agent calls is served by ZeroMind, which has no browser tab and no engine process on your machine: world.open and world.launch answer where a world opens, and this command is what opens it here. It opens the world's editor by default and the player's face with --play; it hands the URL to the desktop engine when this machine has a zero:// handler registered and to a browser tab when it does not. --native / --browser decide it outright and are remembered in open.json beside session.json, so later calls with neither flag follow what you chose. --dry-run prints the target and the URL and changes nothing.
A guid or a URL is opened without a single network call. A name is resolved against your own worlds over /mcp, with this machine's install secret โ which is never printed. No engine is contacted at all.
Two environment variables move the whole CLI to another backend:
| Variable | Default | What it controls |
|---|---|---|
ZEROMIND_ISSUER | https://origozero.ai | Base for every REST call (/v1/installs/*, /v1/worlds, โฆ) and the /mcp URL written into a harness's config. Also the origin of the /link approval page. |
ZEROMIND_CONFIG_DIR | ~/.config/zero (XDG; %APPDATA%\zero on Windows) | Where session.json โ the install_id / install_secret this machine shares with the Zero engine โ lives. Point it elsewhere (e.g. ~/.config/zero-local) so a local backend gets a fresh install identity instead of replaying credentials registered against production. |
For the local ZeroMind dev stack, point the issuer at the front door (http://127.0.0.1:3003), not the bare API on :3001 โ the front door serves the /v1 REST surface, /mcp, and the web app (including /link and the /edit/<guid> engine pages) on one origin, which is what the CLI assumes.
Factual signals from GitHub, npm, and our automated checks โ not a rating.
No reviews yet โ be the first to share how this listing worked for you.
Showcase your server listing on GitHub or your project documentation. Embed this dynamic SVG badge to highlight official listing status and live engagement.
[](https://allmcps.com/mcp/zeromind-origozero-game-engine)<a href="https://allmcps.com/mcp/zeromind-origozero-game-engine"><img src="https://allmcps.com/api/badge/zeromind-origozero-game-engine?style=directory" alt="ZeroMind (OrigoZero game engine) on AllMCPs" /></a>