matchwire MCP for Agents
Mapped market data exposed as tools an agent can call.
Endpoint
POST https://api.matchwire.win/mcp, streamable HTTP, JSON-RPC 2.0, with your key in the same x-api-key header the HTTP API uses. Send Accept: text/event-stream and the single response comes back as one SSE frame; otherwise it is JSON. GET /mcp answers 405 with allow: POST.
The six tools
| Tool | What it returns |
|---|---|
list_sports | Sport names with game counts. |
find_rows | Rows for a sport, venue, type, league or status, or only what changed since a seq. Paged for an AI: 25 rows a call by default (limit up to 200, then offset = the reply's nextOffset), each row with its winner listings and a market count. Add markets: true for every bet type. For bulk use in code, call GET /api/v1/rows. |
get_row | One game by its mapped id, with every matched bet type. Pass type (e.g. winner,total) to keep a props-heavy game small. |
get_venue_market | The venue's own market id and the exact request that reads it — your agent sends that call itself. |
report_issue | Tell us something looks wrong, in plain words: an unmatched or wrongly matched game, a missing game, bad data. Only text is required. A person reads every report; it changes no data. |
service_status | Last cycle, current seq and per-venue health. |
Rules that still hold
Mapping only: nothing through MCP changes data, reloads, or returns a price (report_issue only files a note for a person). Every tool result is narrowed by your key's plan exactly like the HTTP API, the same 5,000-row cap applies, and MCP calls count against the same 60 requests a minute. A key missing or outside its plan comes back as a tool result with isError, naming the limit, rather than as a protocol error.
Clients
Point any MCP client at the URL above with the key as a header. If you would rather not run an MCP client, llms.txt carries the same contract as plain text for a coding assistant.
Parent: docs. Start: quickstart.