sportiq-mcp

MCP server exposing AI-callable tools across FIFA World Cup 2026 football, Formula 1, and IPL cricket.

SportIQ running live in Claude and ChatGPT — Monte Carlo World Cup bracket, F1 pit strategy, and Dream11 optimisation, each backed by a visible MCP tool call. (full 1-min demo)

Three flagship intelligence tools sit on top of raw-data primitives:

• football_simulate_bracket — Monte Carlo with Poisson xG projects World Cup qualification probabilities.

• f1_predict_pit_strategy — tyre-degradation model on OpenF1 telemetry recommends stop laps and compounds.

• cricket_build_dream11_team — PuLP constraint solver picks a valid 11 under credit/role/team caps.

Try it now, no install: a public instance is live on Cloud Run. Add https://sportiq-mcp-329580761892.us-central1.run.app/mcp as a custom connector in claude.ai or ChatGPT — see Use the hosted SportIQ. Open source, read-only, no data collection — why it's safe.

Status

44 tools live: 7 football RAW + 8 football INTEL + 6 F1 RAW + 7 F1 INTEL + 6 cricket RAW + 8 cricket INTEL + 1 cross-sport + sportiq_health. All three flagships shipped: football_simulate_bracket (Monte Carlo + Poisson xG over the 48-team WC 2026 format), f1_predict_pit_strategy (tyre-degradation on OpenF1 telemetry), and cricket_build_dream11_team (PuLP ILP).

Related MCP server: wc26-mcp

Football tools (FIFA World Cup 2026)

RAW

INTEL

The 2026 format (48 teams, 12 groups, top 2 + 8 best thirds → 32-team knockout) is encoded in wc2026.json. Data sources: API-Football (APIFOOTBALL_KEY) → football-data.org (free, token optional) → bundled wc2026.json seed.

F1 Tools

RAW

INTEL

Data sources: OpenF1 (free, keyless) → Jolpica → fastf1 (optional, offline, pip install sportiq-mcp[f1]).

Cricket tools

RAW

INTEL

The Dream11 solver uses CBC via PuLP. On macOS arm64 install with brew install cbc; the binary bundled with PuLP is x86-only and won't run on Apple Silicon.

Cross-sport tools

Diagnostics

Cricket adapter defaults

By default only CricAPI (key required) and static data are active. Opt-in adapters:

SPORTIQ_ENABLE_NDTV=1         # NDTV Sports scraper (operator accepts ToS risk)
SPORTIQ_ENABLE_CRICBUZZ=1     # Cricbuzz scraper (operator accepts ToS risk)
RAPIDAPI_KEY=your_key         # Licensed Cricbuzz mirror via RapidAPI

Copy .env.example to .env and fill in keys.

RapidAPI Hub MCP servers

.mcp.json also wires three external RapidAPI Hub MCP servers (Sportspage Feeds, Football Prediction, Live Sports Odds) via mcp-remote. Because .mcp.json is committed, the API key is a placeholder — replace each <RAPIDAPI_KEY> in .mcp.json with your real RapidAPI key locally to enable them. They run as separate MCP servers and do not affect the in-process sportiq tools.

Install

# from PyPI
uvx sportiq-mcp

# from source
git clone https://github.com/Ninjabeam20/SportIQ-MCP
cd sportiq-mcp
uv sync
uv run python -m sportiq.server

Claude Desktop config

{
  "mcpServers": {
    "sportiq": {
      "command": "uvx",
      "args": ["sportiq-mcp"],
      "env": {
        "CRICAPI_KEY": "your_cricapi_key",
        "APIFOOTBALL_KEY": "your_apifootball_key",
        "THEODDS_KEY": "your_theodds_key"
      }
    }
  }
}

All env vars are optional — the server boots and serves seed/free-source data without any keys. Add a key to unlock the source it gates (e.g. THEODDS_KEY for the value-bet tools). F1 and most football tools use free, keyless sources.

Use the hosted SportIQ (no install — works on claude.ai web & ChatGPT)

A public instance is already running on Google Cloud Run. Add this URL as a custom connector and SportIQ shows up in your AI's tool list — nothing to install:

https://sportiq-mcp-329580761892.us-central1.run.app/mcp

The hosted instance runs without any API keys, so the keyless tools work out of the box: World Cup bracket/group simulations, F1 strategy & tyre models, Dream11 optimisation, match predictions, standings, and schedules. Live-score and live-odds tools (which need rate-limited paid keys) are off on the shared instance — self-host with your own keys if you need those (see below).

Add to Claude (easiest)

1. claude.ai (web): Settings → Connectors → Add custom connector.

2. Name it SportIQ and paste the URL above. Save — the tools appear immediately.

3. Claude Desktop: same path (Settings → Connectors → Add custom connector), or use the uvx config below to run it locally.

Add to ChatGPT

ChatGPT needs Developer Mode turned on first:

1. Settings → Apps & Connectors → Advanced settings → enable Developer mode.

2. In Settings, make sure "use connected apps" (the connectors/tools toggle) is enabled so the model is allowed to call them.

3. Back in Apps & Connectors → Create / Add app (MCP) → paste the URL above, give it the name SportIQ, and connect.

4. Once it shows Connected, start a chat and ask something like "Use SportIQ to simulate the World Cup 2026 bracket" — ChatGPT will call the tools.

First request after an idle period takes ~5–10s (the server scales to zero when unused, so it has to wake up). After that it's fast.

Is it safe to use?

Yes — and here's exactly why, so you can verify rather than take our word for it:

• Completely open source, MIT licensed. Every line is on GitHub and the package is published on PyPI with signed build attestations. Read the code before you connect it.

• Independently reviewed by AI code-audit agents before launch — a full MCP-rubric audit (verdict: ship-ready, no security findings, no secret leak) plus a multi-agent secret/code sweep (verdict: clean). The findings are written up in SECURITY.md so you can check them — and re-run your own audit, since the whole codebase is public.

• Read-only. The tools only fetch and analyse public sports data. There are no write, delete, payment, email, or file-system tools — nothing that can change anything on your side.

• No data collection. SportIQ doesn't ask for, store, or transmit your personal data, prompts, or account info. It answers a tool call and forgets it.

• The hosted instance holds no secrets. It runs with zero API keys, so there's nothing for anyone to steal and no quota of yours to burn.

• Hardened. Upstream content is treated as data (never instructions), API keys are redacted from all logs, payloads are size-capped, and scrapers are opt-in only. See SECURITY.md for the full trust model.

Is the data fresh? Yes. Live sources are polled continuously and cached with tight freshness windows — live scores refresh every ~30s, F1 telemetry every ~10s, standings every ~10min, fixtures every ~6h. Every response carries a meta.is_stale flag and a data age, so the AI tells you exactly how fresh each answer is (e.g. "as of about 4 minutes ago…") instead of guessing. Caching protects free-tier quotas — it never serves you knowingly outdated data without flagging it.

Self-host (your own instance, with live keys)

Prefer to run your own? Set SPORTIQ_TRANSPORT=http and the server serves the MCP endpoint at /mcp (binds 0.0.0.0:$PORT). A ready-to-build Dockerfile is included. See cloud.md for a step-by-step Google Cloud Run deploy (free tier), then add your own https://…/mcp URL as a connector. With your own keys set as env vars, the live-score and odds tools come online too.

Environment variables

Transport: stdio by default (local subprocess — the right fit for Claude Desktop, Cursor, and IDEs). Set SPORTIQ_TRANSPORT=http to serve the streamable-HTTP endpoint at /mcp for remote/web clients (the hosted instance above runs in this mode).

Develop

uv sync --extra dev
uv run pytest
uv run ruff check .
npx @modelcontextprotocol/inspector uv run python -m sportiq.server

See CLAUDE.md for collaboration rules and docs/index.md for the wiki entry point.

Data sources & credits

SportIQ derives some model constants offline from open datasets. Raw datasets are never shipped or fetched at runtime — only small derived seeds (circuits.json, venues.json, elo_seed.json) are committed.

• F1DB — Formula 1 database (1950–present), licensed CC BY 4.0. Used offline to derive per-circuit stop counts and lap lengths in f1/data/circuits.json; per-circuit pit loss is measured offline from OpenF1 lap data (in-lap + out-lap vs clean-lap baseline).

• Cricsheet — free ball-by-ball IPL match data. Used offline to derive measured venue scoring priors (cricket/data/venues.json); we ship only derived aggregates, never the raw match data.

• martj42 international football results — match results 1872–present, CC0. Used offline for Elo backtesting.

• OpenF1 — free, keyless live F1 telemetry (runtime source).

• football-data.org — free football data; their free tier requests a credit link (runtime source).

License & author

Created and maintained by Utkarsh Gupta (@Ninjabeam20).

Licensed under the MIT License — © 2026 Utkarsh Gupta. You may use, copy, and modify this software, but the copyright notice and this permission must be retained in all copies or substantial portions. The canonical package is sportiq-mcp on PyPI and io.github.Ninjabeam20/sportiq-mcp in the official MCP registry.