Using Claude to analyze your hockey data
How an LLM turns your data stack into a question-and-answer machine — safely.
Your question
- Plain language
- "Who are our best PK forwards?"
Schema + tools
- Table definitions
- SQL tool
- MCP server
Claude
- Plans the query
- Calls the SQL tool
- Reads results
Validation
- Check the query
- Sanity-check output
- Human review
Answer
- Numbers + context
- A shareable summary
The 30-second version
- The LLM does not replace your stack — it sits on top of your gold tables.
- Give it tools (SQL, your schema) instead of pasting raw numbers into a chat.
- Keep humans on framing and validation; let the model do retrieval and drafting.
Give the model tools, not a data dump
The biggest mistake is pasting rows of stats into a chat window. Instead, expose your gold tables through a tool (a SQL function or an MCP server) and share the schema. Now the model plans a query, runs it, and reads the real result — grounded in your data, not its imagination.
Let it plan, but keep a guardrail
Claude is strong at translating a plain-language question into a query and explaining the result. Wrap that with a guardrail step: log the generated SQL, sanity-check the numbers against a known total, and keep a human in the loop for anything that drives a real decision.
Turn repeated questions into skills
When the same question comes up weekly — opponent prep, a scouting profile, a post-game summary — capture it as a reusable "skill": a documented prompt plus the tools it needs. That is how a single analyst starts to feel like a department.
Build it yourself
- 1Expose your gold tables via a read-only SQL tool or MCP server.
- 2Paste your table schema into the system prompt so the model knows the shape.
- 3Ask a question and inspect the SQL it writes before trusting the answer.
- 4Save the working prompt + tool set as a named skill for next time.
Newsletter
Weekly hockey analytics
Data-driven takes on performance, scouting, and team strategy. No fluff — just the numbers that matter.
No spam. Unsubscribe anytime.