MCP Access

MCP endpoint

Transport is MCP Streamable HTTP (single JSON-RPC 2.0 endpoint). Auth is shared with REST: pass X-API-Key in the HTTP header; quota, rate limiting and billing are identical.

https://api.sports.csap.ai/mcp (HTTP · X-API-Key)

Claude Code (CLI)

One command adds this platform as an HTTP MCP server — replace YOUR_KEY with your API key:

claude mcp add --transport http football-data https://api.sports.csap.ai/mcp \--header "X-API-Key: YOUR_KEY"

Claude Desktop / Cursor and other clients

Add the following to your client's MCP config file (e.g. mcp.json), replacing YOUR_KEY with your API key:

{ "mcpServers": { "football-data": { "type": "http", "url": "https://api.sports.csap.ai/mcp", "headers": { "X-API-Key": "YOUR_KEY" } } } }

Tools provided

The default compact mode covers every endpoint with 4 generic tools (the AI fetches in three automatic steps); set MCP_EXPOSE_ALL_TOOLS=1 on the server to also generate a first-class named tool per endpoint.

• list_interfaces — list/search the full endpoint catalog (short name, summary, tags, field count)
• describe_interface — return an endpoint's request params and hierarchical field tree
• query_interface — query real data for an endpoint, with id cursor paging / time incremental, **plus flexible entity filtering by team_id/player_id/match_id/competition/season_id/date/live, freely combined** (filters supports any real column)
• live_now — return the current in-play scoreboard in one call (filtered server-side, optional competition filter)
• platform_info — platform overview, billing, paging and realtime access notes

Resources & prompts

Each endpoint's field docs are also exposed as an MCP resource (interface://SHORT-NAME, browsable without entering context); two guiding prompts explore_catalog / incremental_sync are provided too.

Call example (JSON-RPC)

MCP clients send/receive JSON-RPC like the below automatically — you normally don't hand-write it; shown here for reference. This example calls live_now for current in-play matches:

{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "live_now", "arguments": { "competition": "Premier" } } }

{ "jsonrpc": "2.0", "id": 1, "result": { "content": [{ "type": "text", "text": "{ \"count\": 1, \"matches\": [{ \"league\": \"Premier League\", \"home_team\": \"Arsenal\", \"home_score\": 2, \"away_score\": 1, \"minute\": \"67'\" } ] }\" }] } }

Three-step fetch by AI

Once connected you don't need to memorise endpoints — just tell the AI what you need and it will automatically: list_interfaces to find → describe_interface to read fields → query_interface to fetch. When you ask which matches are live now, it uses live_now to get the current scoreboard in one call.

Realtime: pull + push, used together

MCP is for the AI to pull within a conversation: live_now returns the current in-play scoreboard in one call (filtered server-side to live matches only). For continuous push (apps, dashboards, long-running services), exchange the same API key for a one-time ticket, open the WebSocket /ws and subscribe by channel, match or league to keep receiving score, event and odds changes. The two are complementary, one key.

// on your server: exchange the key for a one-time ticket (valid 60 s, single use)const r = await fetch("https://api.sports.csap.ai/api/v1/realtime/ticket", { method: "POST", headers: { "X-API-Key": "YOUR_KEY" } }) const { data } = await r.json() // initial subscription in the URL; change it later with set / subscribeconst ws = new WebSocket("wss://api.sports.csap.ai" + data.ws_url + "&channels=match,incident") ws.onmessage = (ev) => console.log("realtime:", JSON.parse(ev.data))

Note: MCP tools cover REST snapshots and historical data, including the live_now current scoreboard; continuous score, event and odds push goes over WebSocket /ws (exchange a ticket first; the key never appears in the URL). Realtime: WebSocket / SSE / MQTT with live test →