convez.ai
MCP ServerDados & AnalyticsIntermediário
PuckAPI

Claude Sports Analytics – Hockey Intelligence Pack

PuckAPI/claude-sports-analytics

28 skills para análise NHL, modelos de betting e research de hóquei

Ver no GitHub
InicialPluginScore · de 0 a 100

Pack com 28 skills prontas para análise de NHL, modelagem de apostas e pesquisa de hóquei. Funciona com o MCP server PuckAPI para acessar dados ao vivo (ou seus próprios arquivos CSV/JSON). Transforma Claude em analista de hóquei completo sem necessidade de instalação de código.

Instalação

$npx convez-ai PuckAPI/claude-sports-analytics

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
4
Forks
2
Acompanhando
0
Saúde
37%
Último commit
há 3 dias
Versão
v1.1.6
Issues abertas
1
Contribuidores
1

Casos de uso

  • Construir modelos de expected goals (xG) a partir de dados de chutes
  • Analisar estatísticas avançadas de times e jogadores (Corsi, Fenwick, RAPM)
  • Detectar value em odds comparando probabilidades do modelo com cotações de múltiplos bookmakers
  • Backtesting de estratégias de apostas em séries temporais de jogos
  • Análise de goleiros e comparação de desempenho entre jogadores

Veredito

Ferramenta robusta e bem documentada para quem quer analytics de hóquei profissional no Claude. As 28 skills cobrem desde exploração de dados até modelagem preditiva e validação temporal — apropriada para analistas sérios ou apostadores que buscam edge baseado em dados. Integração com PuckAPI opcional, funciona com dados locais também.

README

A tradução para o português deste README ainda está na fila.

Turn Claude into a hockey analyst. 28 free skills covering NHL analytics,betting methodology, and predictive modeling. No code to install.

License: MIT Skills MCP Tools

Quick Start · All Skills · How It Works · Data · Docs

You:    "Build me an expected goals model from shot data"
Claude: [loads xg-model-building skill → walks you through feature selection,
         spatial coordinates, shot type encoding, logistic regression baseline,
         gradient boosting upgrade, calibration, and evaluation]
You:    "What's the edge on tonight's Sabres game?"
Claude: [loads odds-explorer + edge-detection → pulls live odds from 15+ books,
         compares to your model's probabilities, calculates EV, sizes the bet]

Quick Start

1. Clone the skills:

git clone https://github.com/PuckAPI/claude-sports-analytics.git

2. Connect live NHL data (optional — 500 free credits):

claude mcp add puckapi \
  --transport streamable-http \
  "https://mcp.puckapi.com/mcp?key=YOUR_API_KEY"

Get your free key at puckapi.com. Skills also work with your own CSV/JSON files — no API key needed.

3. Ask Claude anything:

"Analyze the Maple Leafs' season — record, Corsi, xG, special teams"
"Help me build a game prediction model"
"Show me tonight's odds and flag any line movement"
"Compare McDavid and MacKinnon across every stat"
"Backtest my betting strategy over the last 3 seasons"

The dispatch skill routes your request to the right 2-3 skills automatically.


Skills

SkillWhat It Does
game-lookupFind games by date, team, season. Scores, schedule, results.
team-analysisStandings, advanced stats, strength of schedule, power rankings.
player-scoutingPlayer search, comparison, NHLe translation, career trajectory.
goalie-analysisGSAA, xSV%, high-danger save rate, workload tracking.
odds-explorerMulti-book odds comparison, best price, line movement detection.
nl-to-queryTranslates natural language questions into structured data queries.
game-previewFull pre-game breakdown: matchup, goalies, trends, odds, prediction.
SkillWhat It Does
hockey-analyticsCorsi, Fenwick, xG, PDO, RAPM, WAR — definitions, context, proper usage.
xg-model-buildingBuild an expected goals model from scratch. Feature engineering through calibration.
SkillWhat It Does
feature-engineeringRolling windows, lag features, leakage detection, feature catalogs.
walk-forward-validationTemporal cross-validation. Refuses k-fold on time-series data.
model-buildingLR → RF → XGBoost ladder. Hard-vote ensemble. Knows when to stop.
elo-engineering5 Elo variants (standard, margin, venue, recency, surface) with tuning.
probability-calibrationPlatt scaling, isotonic regression, reliability diagrams, Brier decomposition.
odds-analysis4 devigging methods: power, multiplicative, Shin, worst-case.
data-pipelineGitHub Actions automation, SQLite schema design, credit budgeting.
SkillWhat It Does
edge-detectionExpected value, Kelly criterion sizing, closing line value tracking.
backtestingWalk-forward historical strategy simulation with realistic constraints.
daily-cardFull slate analysis with edge rankings. Requires your model's probabilities.
bet-trackerLog predictions, track P&L, CLV, and test for statistical significance.
visualizationCalibration plots, edge distributions, player cards, correlation matrices.
SkillWhat It Does
totals-modelingOver/under prediction. Pace metrics, Poisson distribution, under bias.
prop-modelingPlayer prop projections. TOI-first architecture, same-game parlay correlation.
war-gar-decompositionRAPM ridge regression, component GAR, contract surplus valuation.
playoff-simulationMonte Carlo bracket simulation. Series pricing, path probabilities.
SkillWhat It Does
dispatchRoutes your request to the right skills automatically. Start here.
ai-hockey-workflow4 patterns for hypothesis testing: explore, model, validate, deploy.
puckapi-toolMCP tool router. 13 endpoints, credit tracking, BYOD data support.

How It Works

Skills are markdown files, not code. Each one teaches Claude a specific domain. When you mention a topic, Claude reads the relevant skill and gains that expertise for the conversation.

What each skill includes:

  • When to use / when NOT to use
  • Step-by-step methodology
  • Python/pandas code templates
  • Anti-patterns it actively prevents
  • Chaining to the next logical skill

What skills prevent:

  • k-fold on time-series data
  • Future data leakage in features
  • Overfitting to small samples
  • Wrong devigging methods
  • Misinterpreting advanced stats

Data

SourceCoverage
PuckAPI MCP22,000+ games · 107,000+ odds records · 3,000+ players · 16 seasons (2008-present)
Your own filesEvery skill accepts CSV/JSON. Bring your own data, no credits needed.

Reference Docs

DocContents
hockey-glossary.mdCorsi, Fenwick, xG, PDO, RAPM, WAR
betting-glossary.mdOdds formats, vig, devigging, Kelly, CLV
tool-routing.mdDecision tree for picking the right data source
season-logic.mdNHL season IDs, game ID format, lockout handling

Contributing

Found an issue with a skill? Have a methodology improvement? PRs welcome.

Skills are markdown files in skills/[name]/SKILL.md. Each skill has a consistent structure — follow the existing patterns.


PuckAPI MCP Server · puckapi.com · API Docs

MIT License

Ver o original no GitHub

Bom para

Relacionados