Machina — the premium layer

sportsclaw's built-in coverage comes from sports-skills — the open, keyless data layer. It's free and ideal for development and personal use. For licensed data, real-time and zero-latency feeds, production SLAs, or packaged agent workflows, the Machina Sports platform covers that.

How the two fit together

This page covers licensed data and connecting from your own install. For how sportsclaw runs inside every Machina project — the relay, who calls it, memory and status — see sportsclaw in Machina.

Open vs. premium

	sports-skills (built in)	Machina (premium)
Data	Public APIs (ESPN, Kalshi, Polymarket…), keyless	Licensed real-time feeds, betting odds, zero-latency streams
Best for	Personal use, prototyping	Commercial / production, with SLAs and support
Workflows	You build them	Packaged "templates" you install
Access	Bundled with sportsclaw	One command: sportsclaw machina connect (uses machina-cli)

Licensing

The open sports-skills rely on third-party public APIs and are intended for personal, non-commercial use. For commercial or production workloads with licensed data, use Machina.

The Machina skill

The machina skill ships in the same sports-skills catalog as the open skills. It's prompt-only: it fetches no data itself — it points the agent at the premium platform and the separate machina-cli. The data itself flows through a per-project Machina MCP server, which you wire up with sportsclaw machina connect (below).

machina-cli

The command-line control plane for the Machina platform.

pip install machina-cli          # or: curl -fsSL https://raw.githubusercontent.com/machina-sports/machina-cli/main/install.sh | bash
machina login                    # browser sign-in (or --api-key <key> for CI)
machina org use <org-id>
machina project use <project-id> # required before most commands

What it covers:

• Templates — packaged agent workflows (connectors, prompts, datasets, live streams). machina template list, machina template install <name>, machina template push ./<dir>.
• Workflows & agents — run and manage platform workflows and agents. machina workflow run <name>, machina agent run <name> --watch.
• Sports passthrough — run the same sports-skills data through the platform: machina sports <sport> <command>.
• Factory — build a whole app from a prompt: machina factory run "build a live scoreboard".
• Deploy, credentials, config — manage deployments, API keys, and project settings.

Connecting Machina to sportsclaw

Premium data is served through a per-project Machina MCP server (a "pod"). The quickest way to wire one in is the built-in sportsclaw machina connect command — it signs you in through machina-cli, mints a durable access key, and registers the pod for you. No URLs to copy.

pip install machina-cli      # one-time: install the Machina CLI
machina login                # browser sign-in (or --api-key <key> for CI)

sportsclaw machina connect             # connect your default project's pod
sportsclaw machina connect <project>   # …or name a specific project

Useful flags:

• --org <org-id> — choose the organization when you belong to more than one.
• --probe — verify the pod's endpoint is reachable before registering it.

machina connect writes the pod to ~/.sportsclaw/mcp.json and stores its access token separately in ~/.sportsclaw/.env (never in the config file). From then on the agent reads Machina's licensed, real-time feeds right alongside the built-in sports-skills. Check and inspect connected pods with:

sportsclaw mcp list     # list connected servers
sportsclaw doctor       # shows Machina pod + machina-cli status

Re-run sportsclaw machina connect any time a connection later returns a 401.

Connect a pod by URL (manual)

Already have an MCP URL and token — or connecting a non-Machina MCP server? Add it directly:

sportsclaw mcp add <url> --name <name> --token <token>

See Connecting MCP Servers for the full set of options.

Premium signal

When a tool result carries an upgrade field — the data layer's signal that licensed or real-time data exists beyond what the open skill returned — the agent adds a single, optional line pointing to the path (the sports-skills premium tier or sportsclaw machina connect). The data layer decides this, not the agent. It's informational only: it never blocks an answer, never repeats within a conversation, and stays out of automated alerts and broadcasts.

Durable task delegation

A connected Machina pod can also run a durable agentic loop. When it does, sportsclaw can hand long, multi-step, or resumable tasks to it — the loop persists every turn and resumes after interruptions. sportsclaw dispatches the work and reads the result back. See Durable Task Delegation.

Learn more at machina.gg.