Developers
Hockey data your agent can actually call
REST for code, MCP for Claude and Cursor — the same 15 tools, the same credits, no scraping and no schema archaeology.
Building anything on hockey data normally starts with a week of plumbing: find a source, write the ingestion, keep it running, and rediscover every quirk the last person hit.
And if you want an assistant to use it, you then wrap the whole thing in a tool layer yourself.
What to build
Point Claude at it and ask in English
One URL in your MCP config and the assistant picks the tool, fills the parameters and reads the response. No client library, no glue.
Build a matchup page
One call gives both teams, the odds and the goalies who started. It is the whole page in a single request rather than four.
Plot a shot map
Coordinates, shot type, strength and shooter for every attempt, already normalised to one end, so the rendering is the only work left.
What it will not do
- Rate limit is 60 requests a minute, dropping to 10 below 100 credits. Credits, not requests, are the real boundary.
- Responses are JSON only. No GraphQL, no websockets, no streaming.
- Four tools return an untyped payload over MCP — get_plays, get_shot_map, get_player_stats and get_team_stats — so structuredContent is not schema-validated for those.
Worth reading first
Start with 500 free credits
No card. REST or MCP.