Turn Claude into a hockey analyst. 28 free skills covering NHL analytics,betting methodology, and predictive modeling. No code to install.
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
| Skill | What It Does |
|---|---|
game-lookup |
Find games by date, team, season. Scores, schedule, results. |
team-analysis |
Standings, advanced stats, strength of schedule, power rankings. |
player-scouting |
Player search, comparison, NHLe translation, career trajectory. |
goalie-analysis |
GSAA, xSV%, high-danger save rate, workload tracking. |
odds-explorer |
Multi-book odds comparison, best price, line movement detection. |
nl-to-query |
Translates natural language questions into structured data queries. |
game-preview |
Full pre-game breakdown: matchup, goalies, trends, odds, prediction. |
| Skill | What It Does |
|---|---|
hockey-analytics |
Corsi, Fenwick, xG, PDO, RAPM, WAR — definitions, context, proper usage. |
xg-model-building |
Build an expected goals model from scratch. Feature engineering through calibration. |
| Skill | What It Does |
|---|---|
feature-engineering |
Rolling windows, lag features, leakage detection, feature catalogs. |
walk-forward-validation |
Temporal cross-validation. Refuses k-fold on time-series data. |
model-building |
LR → RF → XGBoost ladder. Hard-vote ensemble. Knows when to stop. |
elo-engineering |
5 Elo variants (standard, margin, venue, recency, surface) with tuning. |
probability-calibration |
Platt scaling, isotonic regression, reliability diagrams, Brier decomposition. |
odds-analysis |
4 devigging methods: power, multiplicative, Shin, worst-case. |
data-pipeline |
GitHub Actions automation, SQLite schema design, credit budgeting. |
| Skill | What It Does |
|---|---|
edge-detection |
Expected value, Kelly criterion sizing, closing line value tracking. |
backtesting |
Walk-forward historical strategy simulation with realistic constraints. |
daily-card |
Full slate analysis with edge rankings. Requires your model's probabilities. |
bet-tracker |
Log predictions, track P&L, CLV, and test for statistical significance. |
visualization |
Calibration plots, edge distributions, player cards, correlation matrices. |
| Skill | What It Does |
|---|---|
totals-modeling |
Over/under prediction. Pace metrics, Poisson distribution, under bias. |
prop-modeling |
Player prop projections. TOI-first architecture, same-game parlay correlation. |
war-gar-decomposition |
RAPM ridge regression, component GAR, contract surplus valuation. |
playoff-simulation |
Monte Carlo bracket simulation. Series pricing, path probabilities. |
| Skill | What It Does |
|---|---|
dispatch |
Routes your request to the right skills automatically. Start here. |
ai-hockey-workflow |
4 patterns for hypothesis testing: explore, model, validate, deploy. |
puckapi-tool |
MCP 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
| Source | Coverage |
|---|---|
| PuckAPI MCP | 22,000+ games · 107,000+ odds records · 3,000+ players · 16 seasons (2008-present) |
| Your own files | Every skill accepts CSV/JSON. Bring your own data, no credits needed. |
Reference Docs
| Doc | Contents |
|---|---|
hockey-glossary.md |
Corsi, Fenwick, xG, PDO, RAPM, WAR |
betting-glossary.md |
Odds formats, vig, devigging, Kelly, CLV |
tool-routing.md |
Decision tree for picking the right data source |
season-logic.md |
NHL 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