Quick start: connect your AI

Gamentic is an MCP server. Point your assistant at one URL with one key, and it gains 66 tools for writing, illustrating, scoring, testing and publishing HTML5 games — while you keep your own model and your own context window.

The two values you need

1. Create an MCP key

1. Sign in

   Go to gamentic.net/account. If you do not have an account yet, create one — the Free plan comes with free credits every 30 days.

2. Open MCP Keys

   In the account sidebar, choose MCP Keys, then + New key.

3. Name it after the client

   Use one key per AI client — claude-desktop-laptop, codex-work. Then you can see which machine is generating, and revoke just that one if you need to.

4. Copy it now

   The key starts with gi_. The table only ever shows a truncated form afterwards, so paste it into your client's config before you leave the page.

2. Add the server to your client

Claude Code

One command:

claude mcp add --transport http gamentic https://mcp.gamentic.net/mcp \
  --header "Authorization: Bearer gi_your_key_here"

Or check it in per project by writing .mcp.json at the repository root, which is useful when a whole team should get the same tools:

{
  "mcpServers": {
    "gamentic": {
      "type": "http",
      "url": "https://mcp.gamentic.net/mcp",
      "headers": {
        "Authorization": "Bearer gi_your_key_here"
      }
    }
  }
}

Claude Desktop

Open Settings → Connectors → Add custom connector and give it the endpoint above. Set the Authorization header to Bearer gi_… in the connector's advanced settings. If your build of Claude Desktop does not offer a header field, use Claude Code or another header-capable client instead — there is no OAuth fallback.

Any other MCP client

Cursor, Windsurf, Codex, Zed and most others accept the same shape: a remote server of type http (sometimes called streamable-http), a URL, and a headers map. Check your client's own MCP documentation for where its config file lives, then use the values from the table at the top of this page.

3. Check that it worked

Start a fresh conversation and ask for something read-only:

List my Gamentic games.

Your assistant should call list_games and come back with your library — an empty list is a pass too, if the account is new. If instead you see an authentication error, work through these in order:

4. Fit your agent's modules

Before your first build, open gamentic.net/start. This is the Agent Lab: eleven modules you wire in or cut out, which change the instructions every connected AI receives. It takes effect immediately — the next build follows the new loadout.

5. Build something

Make me a bubble shooter. Cute pastel art, about 3000 credits of budget.
Tell me the plan before you spend anything.

A well-behaved run then goes roughly like this:

1. It reads the rules first

   get_convention and get_skill return the marker convention, the balance guidance and — for a known genre — a gameplay blueprint taken from a real shipped game. This is free and it is what stops the AI guessing at mechanics.

2. It prices the work and asks

   With Budget mode on, you see the scope and the subtotal before any credits move.

3. It builds shell-first

   create_game lands a small game.html shell, then write_game_file fills in one file per character, scene and system. The real headless-browser test runs when the last file lands.

4. It generates and wires the art

   generate_asset, character_sprite, generate_sound, generate_music. The bytes go straight into your game; only names come back to the model.

5. It hands you an editor link

   https://gamentic.net/edit/<id>. From here the tuning is yours — see The Editor.

Next steps