Consume
Bulk export and AI agents
A competition as one Parquet file for your notebook, and an MCP server so an agent can ask the hub directly.
Parquet export
Analytics wants a table, not a paging loop. GET /v1/competitions/{id}/export/fixtures.parquet returns every canonical fixture in the competition as one flat Parquet file — teams, venue, schedule and result columns — readable by pandas, DuckDB, Spark or anything else that speaks Parquet. Secret keys only: bulk is not a widget concern.
The MCP server
@sidelineexchange/mcp (Apache-2.0, like the spec and SDKs) exposes read-only tools over the Model Context Protocol: search teams by name, list fixtures, read one with provenance, every reported result, box scores, standings with their tiebreak traces. It is a thin client of the public API using your own key — its authority is exactly the key’s, so a publishable key gives an agent the public tier and nothing more.
Team search (GET /v1/teams?q=) is the entry point: a name in, canonical ids out. It requires at least two characters — it is a lookup, not a listing.