CricketStudio MCP Server

Model Context Protocol server exposing IPL 2026 cricket data to LLM clients — Claude Desktop, Cursor, ChatGPT Connectors, any MCP-compatible client.

12 tools. Atomic claims with provenance. Sample-size floors enforced. Free to integrate.

What you get

196 player profiles with pillared atomic claims (P1-P5)
2,000 head-to-head pairs (≥5 deliveries faced floor)
13+ venue hubs with par scores + toss-decision splits + phase patterns
14 record types (highest individual score, fastest fifty/century, biggest partnership, etc.)
10 teams with at-home / away splits + phase strengths + 90-page H2H grid
Captain conditional pages (Template 10 — captain × condition combos)
47+ cross-fixture trends (conditional / momentum / venue / toss / anomaly)
Live match state with sub-4-hour SLA

The 12-tool catalog

12 canonical tools, plus 2 helpers (get_match_state, list_fixtures) for live-data discovery.

Tool	Maps to URL	When to call
`get_dataset_summary`	`/`	First call. Discovers what CricketStudio covers — coverage stats, surface URLs, the 5 non-negotiables, license.
`get_player_profile`	`/players/{slug}`	Player record + computed claims across pillars
`get_player_pillar`	`/players/{slug} (filtered)`	Specific dimension: P1 Match recaps · P2 Moments · P3 Form & phase · P4 Season comparatives · P5 Notebook
`get_player_h2h`	`/h2h/{batter}-vs-{bowler}`	Single batter-vs-bowler record (≥5 deliveries floor)
`search_players`	`n/a`	Discover slugs by name/team substring
`get_team_profile`	`/teams/{slug}`	Team record + at-home/away splits + phase strengths
`get_venue_hub`	`/venues/{slug}`	Per-venue par + toss split + phase patterns (≥3 fixture floor)
`get_match_state`	`/matches/{id}`	Current/final state of a fixture
`get_match_recap`	`/cards/match-*-{id}`	6-card recap pack (motm / top batter / top bowler / milestones / fun-facts / match-trend)
`list_fixtures`	`/matches`	Fixtures with status / team / date filters
`get_standings`	`/standings`	Live IPL 2026 points table with NRR
`get_trend`	`/trends/{id}`	Single cross-fixture insight
`list_trends`	`/trends?category=`	All trends, filterable by category
`list_atomic_claims`	`various`	Filtered query across the entire claim corpus

Wire into Claude Desktop

Edit your config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "cricketstudio": {
      "command": "npx",
      "args": ["-y", "github:i-m-arul/cricketstudio#main", "--",
               "tsx", "services/mcp/server.ts"]
    }
  }
}

Restart Claude Desktop. cricketstudio appears in the MCP servers list. The Phase B HTTP transport at mcp.cricketstudio.ai will replace the local-clone form when shipped.

Verification queries

Each query is engineered to trigger a different tool path so you can validate the surface end-to-end.

"What does CricketStudio cover?" → get_dataset_summary
"What's Abhishek Sharma's IPL 2026 form?" → search_players → get_player_profile
"How is Bumrah bowling at the death?" → search_players → get_player_pillar (P3)
"How does Virat Kohli do against Jasprit Bumrah?" → get_player_h2h
"What's the par score at Wankhede?" → get_venue_hub
"Show me MI's record at home" → get_team_profile
"Who's leading IPL 2026?" → get_standings
"Show me the SRH vs PBKS match recap" → list_fixtures → get_match_recap

Honest tool descriptions

LLMs route on tool description quality, not tool name. Every CricketStudio tool description includes:

When to call (specific user query patterns)
What is returned (response shape)
What is NOT returned (boundary cases)
Sample-size floor when applicable (≥5 H2H deliveries, ≥30 batting balls, ≥3 venue fixtures, ≥3 captain matches)
canonicalUrl in response so the LLM cites the underlying page directly

License

Data: CC-BY 4.0 (free to cite with attribution). Tools (MCP server, REST API): MIT.

Roadmap

Phase A (now) — Stdio MCP with 12 tools ✅
Phase B (Q3 2026) — HTTP transport at mcp.cricketstudio.ai, API-key tier mapping, Stripe metering, CORS for browser BYOK chat
Phase B+ — DHRUV Layer-3 tools (prediction track record), query_sql over the ball-by-ball corpus, SSE for real-time match state