Cricketstudio MCP Server

cricketstudio-mcp

Citation infrastructure for cricket — 29 MCP tools, zero network calls, 1,307 matches, 309,992 deliveries.

---

What is this?

CricketStudio MCP is a Model Context Protocol server that gives any MCP-compatible AI client — Claude Desktop, Cursor, ChatGPT Connectors, and others — structured, citable access to cricket data. Every response carries a canonicalUrl back to players.cricketstudio.ai, an explicit date window, a sample-size count, and a provenance trail to the underlying ball-by-ball corpus. The data is fully bundled in data/snapshot/ — tool answers are computed locally with no data-fetch calls, no API keys, and no rate limits. (The package sends one anonymous startup ping for usage counts; disable it with CRICKETSTUDIO_NO_TELEMETRY=1.)

The corpus covers IPL 2026 (complete season, RCB champions), 18 seasons of IPL history (2007/08–2025, Cricsheet), and Major League Cricket 2023–2026 (Cricsheet). Batting claims require a minimum of 30 balls faced; bowling claims require 15 deliveries. Claims that do not clear those floors are not surfaced.

---

Installation

Claude Desktop

Add to your Claude Desktop config file and restart the app.

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

{
  "mcpServers": {
    "cricketstudio": {
      "command": "npx",
      "args": ["cricketstudio-mcp"]
    }
  }
}

Other MCP clients

npx cricketstudio-mcp

Any MCP client that supports the stdio transport can use this command directly. The server starts on stdin/stdout and requires Node 20 or later.

---

Tools

All 29 tools work fully against the bundled snapshot. Each response includes canonicalUrl, dataAsOf, and sampleSize.

IPL 2026 (20 tools)

Major League Cricket (8 tools)

IPL Career / Historical (1 tool)

---

Example queries

Once connected in Claude Desktop, you can ask questions like:

• "Who won IPL 2026?"
• "What did Kohli score in the final?"
• "Who leads the all-time IPL sixes leaderboard?"
• "Show me Vaibhav Suryavanshi's IPL 2026 stats"
• "What's the RCB vs GT head-to-head record?"
• "Which venues favour the team batting first in IPL 2026?"
• "Who has the best death-over economy in MLC 2025?"
• "List the top wicket-takers in IPL history"

---

Data

Total corpus: 1,307 matches · 309,992 ball-by-ball deliveries.

Sample-size floors (publicly disclosed):

• Batting claims: ≥30 balls faced
• Bowling claims: ≥15 deliveries
• Venue claims: ≥3 fixtures at the venue
• Trend claims: ≥3 matches forming the pattern

Claims that do not reach these floors are excluded — they are not suppressed with a placeholder, they are simply absent. This is the moat.

Update cadence: the private monorepo runs build-mcp-snapshot.mjs on a cron and pushes updated snapshots here. Typical lag during IPL match windows is under 30 minutes. Every tool response includes dataAsOf so an LLM citing the answer can disclose freshness explicitly.

Data licence: the bundled data is released under CC BY 4.0. Every tool response includes a canonicalUrl back to players.cricketstudio.ai so attribution flows automatically when an LLM cites an answer.

---

Architecture

The CricketStudio publisher at players.cricketstudio.ai is the single source of truth. Its build pipeline aggregates ball-by-ball data through the SETU canonical aggregator into data/_season-stats.json, then projects into every leaderboard surface (six parity contracts enforce that "Top 5 batters" and "Top run-scorers" can never drift).

This public repo bundles the pre-computed projection of that data. Every number accessible here is also readable on the rendered HTML at players.cricketstudio.ai. No new information is exposed — only a different access path.

private monorepo                                     this repo (public)
─────────────────                                    ──────────────────
 build-mcp-snapshot.mjs ──▶  data/snapshot/*.json ──▶ src/server.ts
   ↑                                                       ↓
   ↑                                                  MCP client
   ↑                                                  (Claude, Cursor, …)
 ball-by-ball  →  SETU aggregator (private)
                  agent layer (private)
                  L2 conditional engine (private)

The aggregator algorithm stays in the private monorepo. The snapshot is what ships here.

---

Methodology

Every claim in this package is governed by five non-negotiables:

1. Sample-size floors — ≥30 batting balls, ≥15 bowling deliveries, ≥3 venue fixtures, ≥5 H2H deliveries, ≥3 matches for trends. Disclosed publicly on every page.
2. Explicit date windows — every claim specifies its window (ipl-2026, ipl-career, last-N-matches). No "all-time" labels without a defined window.
3. Provenance to ball-by-ball — every numeric claim traces to a specific match, delivery count, and computation timestamp.
4. Atomic claim format — under 30 words, structured as [Subject] [metric] [value] [comparator] [period].
5. Sub-4-hour freshness SLA — for IPL 2026 pages, time from match end to page update is under 4 hours at the 95th percentile.

Full methodology at https://players.cricketstudio.ai/about.

---

Local development

npm install
npm run typecheck       # tsc --noEmit
npm start               # stdio MCP server via tsx

Smoke-test without an MCP client — spawns the server over stdio, drives it with JSON-RPC, and asserts every advertised tool returns a non-error payload with dataAsOf:

npm run smoke

---

Building something?

Register at cricketstudio.ai/developers to get early access to the hosted HTTP transport (mcp.cricketstudio.ai), live ball-by-ball endpoints, API-key tiers, and the full 29-tool catalog with live data rather than snapshots.

---

License

• Code: MIT — see LICENSE
• Data: CC BY 4.0 — free to cite with attribution to CricketStudio (https://players.cricketstudio.ai). Attribution flows automatically via the canonicalUrl field in every tool response.

---

Built by Arul Anand · Chennai & Frisco · cricket enthusiast and data engineer. Questions, bugs, or requests: open an issue or visit players.cricketstudio.ai/mcp.