Skip to content
PuckAPI

MCP server

NHL data for AI agents, over MCP

PuckAPI is a hosted MCP server: one URL gives Claude, ChatGPT, Cursor or your own agent 28 read-only hockey data tools. 500 free credits every month, no card.

Get a free keyTry a call in the browser

Connect in one step

Claude and ChatGPT: no key needed

Add a custom connector with this URL. The first time a tool runs, you sign in to your PuckAPI account and press Allow; new users can create a free account from the same screen. In Claude: Settings, Connectors, Add custom connector. In ChatGPT: follow the ChatGPT steps below.

url
https://mcp.puckapi.com/mcp

Claude Code

bash
claude mcp add --transport http puckapi https://mcp.puckapi.com/mcp

Then run /mcp in Claude Code and sign in. To use an API key instead, add ?key=YOUR_API_KEY to the URL.

Claude Desktop, Cursor and other clients

json
{
  "mcpServers": {
    "puckapi": {
      "url": "https://mcp.puckapi.com/mcp",
      "headers": { "x-api-key": "YOUR_API_KEY" }
    }
  }
}

Listing tools works without signing in; calling one needs your account or an API key, and uses its credits.

Use it in ChatGPT

ChatGPT adds outside MCP servers through developer mode. OpenAI currently offers it on workspace plans (Business, Enterprise and Edu), where an admin may need to turn it on, and it moves its menus often. Its developer mode guide has the current location of each setting.

  1. Turn on developer mode in ChatGPT's settings (or ask your workspace admin to).
  2. Create a custom MCP app or connector. Name it PuckAPI, paste the server URL below, and choose OAuth as the authentication.
    url
    https://mcp.puckapi.com/mcp
  3. ChatGPT opens a PuckAPI sign-in page. Sign in, or create a free account on the same page, and press Allow. There is no API key to copy, so none ends up in a chat.
  4. Start a new chat, add PuckAPI from the tools menu, and ask something like “Which goalies have saved the most goals above expected this season? Show the top 10 with games played.”

No developer mode on your plan? Use Claude (steps above), or call the REST API from your own computer: your key stays on your machine and the responses save straight to files. See save the raw JSON in the docs.

Other assistants and editors

The same URL works anywhere that takes a remote MCP server. Where the app signs you in, use your PuckAPI account; where it asks for headers, send your key as x-api-key.

Gemini CLI

bash
gemini extensions install https://github.com/PuckAPI/mcp

Then set PUCKAPI_API_KEY to your key before starting Gemini CLI.

VS Code

json
// .vscode/mcp.json
{
  "servers": {
    "puckapi": { "type": "http", "url": "https://mcp.puckapi.com/mcp" }
  }
}

Zed

json
// settings.json
{
  "context_servers": {
    "puckapi": {
      "url": "https://mcp.puckapi.com/mcp",
      "headers": { "x-api-key": "YOUR_API_KEY" }
    }
  }
}

Or leave out headers and Zed signs you in to PuckAPI instead.

Google Sheets

Copy the template sheet, add your key under Extensions, then type =SDHQ("puckapi","get_standings") for a live table.

Grok

At grok.com/connectors choose New connector, then Custom, paste the server URL and sign in to PuckAPI.

Perplexity and Le Chat

Both take a custom remote connector: Perplexity on Pro, Max and Enterprise plans, Le Chat from the account owner's connector settings. Paste the server URL and sign in, or add your key as an API key.

Ask it things like

28 tools

Credits per call. The same tools are REST endpoints; see the docs.

All 32 active NHL teams with divisions, conferences, and arenas

Current standings with advanced stats

Upcoming games for next N days

Find players by name, team, or position

Query by date, team, season, game state, or type

Detailed stats for a specific team with advanced metrics

Player bio and team context; season line for skaters, goalie stats for goalies

Goalie leaderboard sorted by save%, GAA, GSAX, or wins

Skater leaderboard sorted by points, goals, assists, +/-, or shots

A player's game-by-game stat lines for a season, not a season total

The model's win probability and A/B/C grade for each game on the current slate, with the model version and a backtest note

Paid plans only. The model's projected goals, assists, points, shots and saves for tonight's players, with ranges and over probabilities

Odds by game with bookmaker filters

Full game detail with odds and goalie starts

Matchup history with win/loss records

Play-by-play events with rink coordinates, strength state, and the players involved

Every shot attempt with rink coordinates, shot type, shooter and goalie

Every player's shifts in a game, to the second, with official time on ice and plus/minus

Exact seconds each team spent at 5v5, 5v4, 4v5, 6v5 and every other on-ice state, by game or date

Time-series odds grouped by bookmaker

Player prop lines by sportsbook: the closing line; the opening line where held

A player's prop lines against the result, game by game

How one player's prop line moved before a game, every capture by sportsbook

Opening and closing lines by sportsbook for the period and regulation-result markets (history from May 2023)

Alternate puck lines and totals by sportsbook: every point on each book's ladder, opening and closing, from 2023-24

What the market thinks of a game: average implied and fair probability for the moneyline, puck lines and totals

What the market thinks of a game's player props: average implied and fair probability for each player and line

How often a kind of player prop hit against what the market implied, by your filters, since May 2023

Questions

What is the PuckAPI MCP server?

A hosted Model Context Protocol server at https://mcp.puckapi.com/mcp. Add that URL to Claude, ChatGPT, Cursor or any MCP client and the assistant can call 28 read-only hockey data tools directly: no scraping, no local install.

Do I need to run anything locally?

No. It is a remote server over Streamable HTTP at one URL. Claude and ChatGPT connect with no key: they ask you to sign in to your PuckAPI account the first time. Other clients send your API key as an x-api-key header or a ?key= parameter.

What does it cost?

Every account gets 500 free credits every month and no card. Connecting from an app starts you with 50 while you confirm your email, which adds the other 450. Tools cost 1 to 25 credits a call, a call that returns no data is refunded, and plans start at $19/month for 10,000 credits.

Is it the same data as the REST API?

Yes. Every MCP tool is also a REST endpoint with the same arguments and response, so an agent prototype and production code read identical data.

Which clients work?

Anything that speaks MCP over Streamable HTTP: Claude Code, Claude Desktop, ChatGPT (connectors), Cursor, VS Code, Zed, Cline, Windsurf and agent frameworks with an MCP client. It is also listed on the official MCP Registry and Glama.

Building analysis on top? The skills library has ready-made Claude workflows for this data.