Point your tools at the platform

This section holds what a developer or an AI agent can point a tool at. Everything in it is fetchable, installable, or callable: the API contract as a file, the same contract as a browsable reference, a typed client on npm, a hosted MCP endpoint, and the whole documentation as plain text. The integration itself is explained in Docs; this section is the machinery you build it with.

Take the API contract before you register

One command, no key, no account:

curl -sS https://hub.aggregator.gg/openapi.yaml | head -4

Expected response:

openapi: 3.1.0
info:
  title: The Aggregator Machine API
  version: '1'

This is the OpenAPI file that describes our production API, ready for a code generator. The API Reference on this hub is rendered from the same file.

What is here

Surface	What it is	Where it starts
OpenAPI spec	The machine API as one YAML file, downloadable without a key.	openapi.yaml
API Reference	The same spec rendered page by page, one page per operation.	API Reference
Node.js SDK	`@aggregator-gg/sdk` on npm: a typed client for games, sessions, and transactions, with typed errors for API and transport failures.	Node.js SDK
Code samples	Four of the five operations every integration needs as code in TypeScript, PHP, and Python, plus the requirements for the callback handler.	Code samples
Remote MCP	A hosted MCP server at `mcp.aggregator.gg` so an agent reads catalog and integration state through typed tools instead of hand-written HTTP.	MCP server
llms.txt	The documentation as machine-readable text: an index of approved pages and a full-text corpus.	llms.txt
Agent guidelines	What an agent may do alone, what stays with a human, and how keys are handled.	Agent guidelines

Two of these deserve the caveat up front. The hosted MCP endpoint is read-only today: its five remote actions stay listed but are denied at call time, so an agent reads state there and prepares integration code; the human runs remote tests and launches through an authorized application using the SDK or HTTP. And the SDK covers the operator side of the API, not the wallet callbacks your own server has to answer; that contract is on Callbacks.

How everything here authenticates

One key covers all of it. You create an agg_ key in the cabinet and send it as an Authorization: Bearer header on every call, to the API and to the MCP endpoint alike; the MCP server forwards the key and does not store it. Nothing on this page needs a key to read: the spec, the reference, and both llms files are public files.

You prove the money path against that same production key: confirmed access and budget enable controlled real rounds and demo sessions launch games with no money involved. Environments explains the model and lists the platform egress IPs to allowlist, and API keys covers creation, rotation, and scope.

If you are the AI agent reading this

Fetch https://hub.aggregator.gg/llms-full.txt once and treat it as the documentation corpus: it is every approved page of this hub as plain markdown, generated from the same registry that renders the site. Then read Agent guidelines before you act on any of it. The short version: the key belongs in your client configuration and never in a prompt or in tool output, and the go-live decision belongs to a human.