Quick Start · MCP + agent skills

Your idea. A playable browser game.

Install the AIGameLabs connection and publishing workflow, then give your agent one complete request. Build and test first, preview privately, and submit for review.

Step 1 · Install

MCP + publishing skill, together

Use Node.js 20 or newer and an already-installed supported agent. Open a terminal in your game project. You will also need an AIGameLabs account and access to your agent's MCP connection settings.

Download to a fresh temporary file and read the script before executing it. Keep the same terminal open for all three steps. Stop if a command fails; do not run the remaining steps. No piping downloaded code into a shell.

Download and inspect the installer

setup_dir="$(mktemp -d "${TMPDIR:-/tmp}/aigamelabs-setup.XXXXXX")" &&
setup_file="$setup_dir/setup.mjs" &&
curl --fail --show-error --location 'https://aigamelabs.net/mcp/setup.mjs' --output "$setup_file" &&
cat "$setup_file"

Preview detected clients — no writes

node "$setup_file" --dry-run

Apply the reviewed setup

node "$setup_file"

By default, the installer detects supported clients and configures their documented files. It merges the AIGameLabs entry without replacing unrelated settings and refuses conflicting existing entries or skill files instead of overwriting them. Read its report for each client and any required manual steps.

Choose clients and control what the installer changes
--list
List the supported client IDs and setup limitations.
--agents codex,claude-code
Select specific clients rather than relying on detection.
--all
Select every listed target, even if it is not installed. UI-managed apps still need manual setup.
--allow-bridges
Opt in to configurations that let your client run mcp-remote later. Bridges are not configured by default and are not run by this script.
--project PATH
Choose the project root for project-scoped configuration and skills. User-level settings remain user-level.
--dry-run
Preview planned changes and conflicts without writing any files.
--skill-file PATH
Use a local, already-reviewed SKILL.md instead of downloading it.
--help
Show all command options.

One workflow · your agent's folder

Shared skills, per-agent connections

The MCP connection

Authenticated tools for drafts, build and artwork uploads, private previews, and releases.

AIGameLabs MCP server

https://aigamelabs.net/mcp

Each client signs in with OAuth. No shared API key or token in configuration.

The publishing skill

The instructions for web-capable builds, safe packaging, accurate metadata, rights, and review.

Publishing skill name

aigamelabs-game-publishing

Read the complete skill

Codex, Cursor, VS Code and several other agents can share a project skill in .agents/skills/aigamelabs-game-publishing/SKILL.md. Claude Code and Cline use their own skill folders; omp also has its native .omp/skills/ path. The installer follows each target's documented location rather than assuming one folder works everywhere. A reviewed project skill can be committed for your team or cloud agent; OAuth credentials must never be committed with it.

Installer coverage · 29 clients

File configuration still needs client-side OAuth. Availability depends on your platform and existing files; the installer report lists skipped or manual work. These are documented setup targets, not a claim that every client has been runtime-tested.

Client guideMCP configurationSkill installation
Oh My Pi (omp)File configurationNative skill file
Claude CodeFile configurationNative skill file
OpenAI CodexFile configurationNative skill file
Gemini CLIFile configurationNative skill file
OpenCodeFile configurationNative skill file
Qwen CodeFile configurationNative skill file
GooseFile configurationNative skill file
AmpFile configurationNative skill file
WarpFile configurationNative skill file
OpenHandsFile configurationNative skill file
Kiro CLIFile configurationNative skill file
CursorFile configurationNative skill file
VS Code (GitHub Copilot)File configurationNative skill file
WindsurfFile configurationNative skill file
ZedFile configurationNative skill file
JetBrains AI AssistantManual setupNative skill file
Kiro IDEFile configurationNative skill file
ClineFile configurationNative skill file
Roo CodeOpt-in bridgeNative skill file
Kilo CodeFile configurationNative skill file
ContinueOpt-in bridgeReference only
Claude (web, Desktop & Cowork)Manual setupManual setup — see guide
ChatGPT (web custom apps)Manual setupReference only
LM StudioManual setupReference only
Cherry StudioManual setupManual setup — see guide
LibreChatManual setupManual setup — see guide
Open WebUIManual setupManual setup — see guide
AnythingLLMOpt-in bridgeReference only
JanManual setupManual setup — see guide

Step 2 · Connect

Restart, sign in, and check the connection

  1. Review the setup report. Restart the configured client or reload its MCP and skill settings as its guide requires.
  2. Use the client's Connect or Authenticate action for aigamelabs. Complete the browser sign-in and inspect the consent screen. Publishing needs games:read and games:write.
  3. Confirm the publishing skill is available. Then ask the agent to list your own game entries without changing anything before giving it a publishing task.
Configuration is not authorization

The installer writes configuration and Markdown only. It does not install clients, run login commands, create an OAuth session, grant consent, or bypass creator access. Never paste tokens, browser cookies, or account credentials into a config file or chat.

UI-managed desktop and web apps need their own manual connector or skill-import steps. A reference-only app must read or attach the Markdown; that is not native skill discovery and does not connect MCP. Follow your client's guide for the exact setup and OAuth flow. AIGameLabs has no API-key fallback.

Step 3 · Create

One prompt, from game idea to review

Replace <GAME> with your idea, then paste this into your connected coding agent.

Create and submit a browser game

Create <GAME>.

Read and follow the aigamelabs-game-publishing skill. Use the authenticated AIGameLabs MCP server at https://aigamelabs.net/mcp for the real publishing workflow, not simulated tool results.

Implement a complete, playable WebGL/browser demo with clear controls, a core gameplay loop, and restart/win/lose states appropriate to the game. Choose a web-capable stack such as a web-native engine, Canvas/WebGL, or a supported Unity/Godot web export. Do not assume stock Unreal Engine 5 can export WebGL; it has no official HTML5/WebGL target. Use original or properly licensed game assets and bundle all runtime dependencies locally. Keep the build single-threaded, with no workers, service workers, or external runtime requests, and follow the skill's sandbox and size limits.

Run the actual production build in a browser, play through the core loop, inspect the console, and fix problems. Package the tested static output as the skill requires: a ZIP with index.html at its root and relative asset paths. Do not upload an untested build.

Use AIGameLabs MCP to create my private game draft, upload the finished build, upload cover artwork and actual gameplay screenshots, and fill in accurate metadata, controls, supported devices, and AI-use disclosures. Open and exercise the private preview returned by preview_build in the real player sandbox; fix issues before requesting review. Do not claim that an upload, preview, or review succeeded without observing its result.

Record the game's actual or estimated creation cost in USD only when the token, tool, and asset costs are known. Clearly distinguish an estimate from a measured total; leave the creation cost unspecified rather than inventing a number. This is the cost to create the game, not a price to play it.

Before asserting rightsAccepted: true, ask me to explicitly confirm that I own or have permission to distribute all code, game assets, artwork, music, and text, and that the AI-use disclosures are accurate. This prompt is not that rights confirmation.

I explicitly authorize autoPublish: true on create_release for the exact tested build and reviewed metadata/artwork, so that release may publish automatically only after moderator approval. After my rights confirmation, freeze that release and use submit_release to send it for human review. Do not bypass moderation or claim immediate public publication.

Give me the private preview link, an honest summary of what you tested, the draft/release status, and any remaining action I must take. If authentication, creator access, rights confirmation, quotas, or moderator feedback blocks a step, explain it and wait for the required action. Never ask me to paste OAuth tokens or account credentials into chat.

Your rights confirmation still comes first

This prompt explicitly permits automatic publication of the exact release after moderator approval. It does not attest that you own the rights: your agent must ask you to confirm them before creating the release. Sign-in, creator access, build fixes, limits, and human review may still require action. One request is not a promise of instant publication or a successful build in one attempt.