User Guide — Battle Court Stats in Claude

Welcome! This connector gives Claude direct, read-only access to the Battle Court (World Jai Alai League) statistics database — every match charted throw by throw since Fall 2024. You ask questions in plain English; Claude looks up the real numbers and answers with them.

No spreadsheets, no SQL, no exports. Just ask.

One-time setup (2 minutes)

1. Open claude.ai → Settings → Connectors (same place in the Claude desktop app).
2. Click Add custom connector and paste the URL you were given: https://mcp.reboteanalytics.com/mcp
3. A Battle Court Stats sign-in page opens. Either:
   • click Continue with Google and pick your account (easiest), or
   • enter your email + password. First time? Open the invite link you were sent and choose a password — you'll be connected in the same step.
4. Back in Claude, make sure the connector is enabled in your chat's tools menu (the sliders icon).

That's it. You'll occasionally be asked to sign in again (roughly monthly if you haven't used it) — same one-click process.

On mobile: set the connector up on the web or desktop app first; it then works in the mobile apps automatically.

How to ask

Just describe what you want to know, the way you'd ask an analyst:

"Who has the best rebote percentage this season among backcourters — minimum 40 attempts?"

Claude figures out which lookups to run, pulls live numbers, and explains the answer. A few habits get you noticeably better results:

Player names resolve automatically — if you type "goenaga" or even misspell it, Claude searches the roster for the right spelling before running stats.

What's in the database (plain English)

• Doubles matches (2v2) from Fall 2024 onward and singles (1v1) from Winter 2025 onward — lineups, ranks, final scores, playoffs flags.
• Point-by-point logs: every point's rally length, server, winner, running score, and how it ended (winner, tip, out, screen, pad, double fault, chula…).
• Throw-by-throw logs: every individual ball touch — who threw, from where on the court, what they threw (placement and side), and what happened to it. This unlocks situational questions: kill rates from the outside of the court, performance against specific opponents, off-the-back-wall throws, "% of throws not returned" in any scenario you can describe.
• Per-player, per-match stats in five families:
  • Catching — catch rates overall and by side (forehand / backhand / costado)
  • Serving — serve %, first-serve %, aces, double faults
  • Kill shots — attempts and success by type (two-wall, splitter, filo, dejada, three-wall, cortada)
  • Rebotes — back-wall catch success (the biggest skill separator in the league)
  • Errors — balls thrown out, into the screen, or into the pad
• Team context — who carries the throwing load on each pair, franchise results, ranks (1 = elite … 6 = bottom).

Anything Claude reports is computed from this real data at the moment you ask. New matches appear after the league's stats pipeline runs — ask "What's the most recent match date in the database?" anytime to see how current it is.

Built-in shortcuts

Two one-click reports live in Claude's prompt menu for this connector:

• Scouting report — a full profile of any player: record, catching, serving, kill shots, rebotes, errors, league context, strengths/weaknesses.
• Compare two players — side-by-side stats with a plain-English verdict.

You can always just ask for these in words too.

Trust & limits

• Read-only, guaranteed. Nothing you or Claude can type changes the database — the connection is locked to reading at four independent layers. Experiment freely; you cannot break anything.
• Counting conventions (worth knowing): a "kill" = a kill-shot attempt that won the point; a double fault counts as two faults (first + second serve); rates are always recomputed from raw counts, never averaged.
• Substitution matches are excluded from player skill stats by default (they distort who-did-what); results listings still include them, labeled.
• Small samples — if a stat is built on 6 attempts, Claude is instructed to say so. Take those numbers as hints, not conclusions.
• Long custom analyses may take a few seconds; very heavy queries are automatically time-capped at 15 seconds.

If something's wrong

Need more? See the FAQ and the sample questions gallery.