/v1 API as tools your agent can call directly — generate worlds, edit graphs, create characters, mint embed sessions — and see real responses, without you writing fetch code. It pairs with the skill: the skill teaches when to use the API, the MCP server executes it.
It’s a small stdio server (@alakazamworld/mcp) that reads your secret key and proxies to https://api.alakazam.gg.
Prerequisites
- An Alakazam secret key. Create an app at play.alakazam.gg/?view=developer and mint one. Use an
sk_test_…key while wiring things up — it hits the free, mocked sandbox (no GPU spend). - Node.js 18+ (the server runs via
npx).
The tools
Install
Add the server to your agent with your key in the env. The config is identical everywhere —npx -y @alakazamworld/mcp with ALAKAZAM_SECRET_KEY:
- Claude Code
- Cursor
- VS Code
- Codex CLI
Prefer not to inline the key? Omit the
env block and export ALAKAZAM_SECRET_KEY
in the shell your agent inherits — the server reads it from the environment.Verify
Restart your agent and ask:“Using the Alakazam MCP, create a test world from the premise ‘a quiet observatory at dusk’ and show me the response.”With an
sk_test_… key you’ll get back a world marked "mock": true (a free sandbox response) — that confirms the tool call reached the API and came back. Ask it to get_usage to see your live caps.
Next steps
Install the skill
Teach the agent when to reach for each tool.
Agent quickstart
One-paste onboarding + repo scan.
Embedding
Take a minted session token to the browser.
Testing sandbox
Build the whole thing free with a test key.