---

name: puckapi-tool description: "Default data source for PuckAPI Skills. Routes all MCP tool calls for NHL games, schedules, team standings, player stats, goalie performance, and betting odds. Use when any other skill needs data. Do not invoke directly -- other skills call this. Not for data analysis, modeling, or bet recommendations -- see hockey-analytics, model-building, or edge-detection instead." metadata: version: 1.0.0 author: PuckAPI user-invocable: false

PuckAPI Tool

Every other skill delegates data retrieval here. This skill owns the 15 MCP tools, routes requests to the right endpoint, tracks credit cost, and handles the fallback when users bring their own data.

Database coverage: 6,882,653 plays | 23,495 games (2010-11 onward) | 107,058 odds records (2019-20 onward, sparse in the first year) | 4,898 players | 1,509 goalie-seasons | 1,803 skater-seasons | 21,994 games with derived goalie starts | 494 standings records | 34 franchises (32 active + 2 historical)

Full parameter schemas: See endpoints.md in this directory (Level 3 -- load on demand).

---

When to Use Each Tool

Task	Tool	Credits
List all teams	list_teams	1
Upcoming schedule	get_schedule	2
Find a player by name	search_players	2
League or conference standings	get_standings	2
Games by date, team, or season	get_games	5
Player bio + stats by NHL player ID	get_player_stats	5
Single team stats + advanced metrics	get_team_stats	5
Goalie leaderboard	get_goalie_stats	5
Full game detail with odds + goalies	get_game_detail	10
Head-to-head history between two teams	get_head_to_head	10
Odds snapshot for a game	get_odds	10
Opening vs closing line movement	get_line_movement	25

---

Tool Selection Guide

Use this table when a request could match multiple tools.

You Need	Use This	Not This
"What games are on tonight?"	get_schedule	get_games (returns past games too)
"Show me BUF's last 10 games"	get_games with team=BUF, limit=10	get_schedule (future only)
"Full game info with odds"	get_game_detail	get_games + get_odds separately
"How have BUF and TOR fared against each other?"	get_head_to_head	get_games filtered manually
"Where does BUF rank in the Atlantic?"	get_standings with division=Atlantic	get_team_stats
"BUF's Corsi and xG numbers"	get_team_stats	get_standings (standings have basic metrics only)
"Find Connor McDavid's ID"	search_players with query=McDavid	get_player_stats (requires numeric ID)
"McDavid's bio and current stats"	get_player_stats with known ID	search_players
"Best save percentage goalies this season"	get_goalie_stats with sort_by=save_pct	get_player_stats
"What's the line on tonight's BUF game?"	get_odds with that game's ID	get_line_movement (movement, not snapshot)
"Did DraftKings move the line today?"	get_line_movement	get_odds

---

Credit Usage

Tier	Endpoints	Credits
Lightweight	list_teams	1
Standard	get_schedule, search_players, get_standings	2
Moderate	get_games, get_player_stats, get_team_stats, get_goalie_stats	5
Heavy	get_game_detail, get_head_to_head, get_odds	10
Expensive	get_line_movement	25

Bulk cost examples:

Operation	Credits	Notes
Full season odds backfill	~12,300	10 credits x ~1,230 games
Full season game details	~12,300	10 credits x ~1,230 games
Playoff odds (full bracket)	~150-250	Depends on rounds reached

Odds coverage caveat: Odds data covers 2020-2026 NHL seasons. Calls to get_odds or get_line_movement for games before the 2019-20 season return empty odds arrays, not an error.

---

Commands That Do NOT Exist

Not Available	Use Instead
get_live_odds	get_odds with today's game ID (snapshot_type omitted returns all)
get_player_advanced_stats	get_player_stats, then compute via hockey-analytics
get_team_schedule	get_games with team=ABBREV
get_game_by_teams	get_head_to_head or get_games with team filter
get_playoff_bracket	get_games with game_type=playoff
get_historical_lineups	Not available -- no lineup data in this dataset
get_goalie_game_log	Not available -- use get_game_detail for per-game goalie starts
get_injuries	Not available
get_power_play_stats	Not available as a standalone endpoint
search_games	Use get_games with filters

---

Season Resolution

Every skill that touches time-sensitive data must resolve season IDs before calling tools.

Season ID format: YYYYYYYY (eight digits, no separator). Example: 2025-26 season = 20252026.

Current date resolution rules:

• October through December: current calendar year starts the season (Oct 2026 = season 20262027)
• January through September: previous calendar year starts the season (Mar 2026 = season 20252026)
• "This season" = season currently in progress or most recently completed
• "Last season" = one season before "this season"

NHL calendar:

• Regular season: early October to mid-April
• Playoffs: mid-April to mid-June
• Offseason: July to September
• Data coverage starts at 20102011 (2010-11 season). There is nothing before it; querying an earlier season returns empty, not an error

How to resolve when user says "current season":

1. Call get_standings with no season parameter -- the server resolves automatically
2. Or call get_team_stats with no season -- same auto-resolution
3. For get_games, pass season=20252026 explicitly when targeting a specific season

---

BYOD (Bring Your Own Data)

When a user provides their own CSV, JSON, or database instead of using PuckAPI:

1. Verify required columns exist before proceeding (see endpoints.md for column schemas)
2. Check date format -- ISO 8601 preferred (YYYY-MM-DD). Flag anything else.
3. Flag missing data explicitly -- do not silently drop rows or impute without noting it
4. No credits consumed when working with user-provided data
5. Column name mapping -- if user data uses different column names, map them to the PuckAPI schema before passing to analytics skills

Common user-provided formats:

Format	Required Minimum Columns	Notes
Game results CSV	game_date, home_team, away_team, home_score, away_score	game_id helpful but not required
Odds CSV	game_id or (game_date + home_team), bookmaker, ml_home, ml_away	Snapshot type required for line movement analysis
Player stats CSV	player_id or player_name, team, season, stat columns	Prefer numeric ID over name for joins

---

How to Call These Tools

Pattern: Find game ID then fetch detail

Most analytics workflows need a game_id. Get it from get_games or get_schedule, then pass to get_game_detail, get_odds, or get_line_movement.

Step 1: get_games(team="BUF", date_from="2026-01-01", date_to="2026-01-31")
        -> returns game list with IDs
Step 2: get_game_detail(game_id="2025020887")
        -> returns full game with odds and goalie starts

Pattern: Player lookup by name then stats

get_player_stats requires a numeric NHL player ID. Always search first.

Step 1: search_players(query="McDavid", active=true)
        -> returns player list with IDs
Step 2: get_player_stats(player_id=8478402)
        -> returns bio + career stats

Pattern: Standings with advanced metrics

get_standings includes Corsi, Fenwick, and xG columns. No need to call get_team_stats separately for a full league view.

get_standings(conference="Eastern", season="20252026")
-> returns all Eastern teams sorted by rank with advanced metrics

---

Anti-patterns

What Claude Might Do	Why It's Wrong	Do This Instead
Call get_odds for every game in a season	10 credits x 1,230 games = 12,300 credits	Warn user before bulk odds pulls, ask for date range
Call get_line_movement when user just wants current odds	25 credits wasted	Use get_odds with snapshot_type=current (10 credits)
Call search_players and then guess the ID	IDs must be exact	Use the actual id field from search_players results
Pass a team name like "Buffalo Sabres" to team parameter	Tool expects abbreviation	Always use 2-3 letter abbreviation (e.g. BUF)
Assume odds exist for pre-2020 games	Returns empty array, not an error	Check game season before calling odds endpoints
Call get_game_detail in a loop for many games	N x 10 credits per game	Use get_games (5 credits) for bulk data, get_game_detail only when you need odds/goalies for a specific game

---

What to Do Next

What You Retrieved	Next Action	Skill
Game results for a season	Compute Elo ratings or features	elo-engineering or feature-engineering
Standings with Corsi/Fenwick	Analyze team performance trends	team-analysis
Odds for upcoming games	Find edge vs market	edge-detection
Line movement data	Detect sharp action signals	odds-analysis
Player/goalie stats	Scout performance vs league	player-scouting or goalie-analysis
Head-to-head history	Build game preview	game-preview
Full season game log	Run a backtest	backtesting
Raw game data, need features	Engineer model inputs	feature-engineering