The open sports intelligence library for AI agents and developers.

SportMind teaches AI agents how to reason about sports — not just react to data.

SportMind is a reasoning library, not a data feed. Every file teaches an agent how to think — not what is true right now. Load a skill, and your agent immediately understands the sport, the athlete, the commercial landscape, and the external forces acting on it.

→ WHO-USES-THIS.md — Developer, agent builder, analyst, researcher, or contributor? This maps you to exactly the files you need.

AI agents that analyse fan tokens, run prediction markets, or power sports GameFi need more than raw data. They need context:

• That a weigh-in miss in MMA is categorically different from a team losing a regular season game
• That a cricket match on a Mumbai evening will be affected by dew in the second innings
• That a DAXA Investment Warning on a Korean exchange is a staged lifecycle event with a predictable intervention window — not a binary delisting signal
• That a new Binance listing aligned with a club's actual fanbase geography extends CDI durability; a misaligned listing does not
• That Galatasaray equity on Borsa Istanbul (GSRAY.IS) leads the GAL fan token by 24–72 hours on commercial news — both instruments are pricing the same underlying entity
• That a liquidity pool with $80k TVL will absorb your signal's value in slippage before you execute

This contextual reasoning is currently rebuilt from scratch by every developer in the space. SportMind is the shared layer.

Option A — Paste into any LLM (zero setup)

1. Open Claude, GPT-4, Gemini, Groq, Mistral, or any LLM
2. Paste: contents of core/sportmind-purpose-and-context.md
3. Paste: contents of sports/football/sport-domain-football.md
4. Ask: "PSG vs Arsenal UCL tonight. PSG full squad. Arsenal striker injured.
         Using SportMind, generate a pre-match signal."

Working in under 3 minutes.

Option B — Skills API

python scripts/sportmind_api.py   # start local API

curl "http://localhost:8080/bundle/ftier1-football"   # named bundle
curl "http://localhost:8080/stack?sport=football&use_case=fan_token_tier1"

Option C — Clone and run

git clone https://github.com/SportMind/SportMind
pip install aiohttp --break-system-packages
python examples/starter-pack/01-simple-signal.py

Load order: macro → market → sport domain → athlete → fan token → output schema

Use a named bundle: ftier1-football · ftier1-cricket · prematch-mma · governance-brief → platform/skill-bundles.md for all 14 bundles with token estimates.

743 files · 522 markdown skill files

Sport domain:      42 sports · event playbooks · risk variables · agent reasoning prompts
Athlete:           29 sports · form models · availability · composite modifier (0.55–1.25×)
Fan token:         85 verified tokens · Lifecycle phases 1–6 · DeFi liquidity · exchange intelligence
                   Fan Token Play PATH_2 · governance · KOL influence · FTO framework
                   Sports equity signals (GSRAY.IS, MANU, JUVE.MI, FWONK, TKO) · CHZ macro layer
Market:            43 documents · club operations · broadcaster intelligence · World Cup 2026
Macro:             25 documents · MiCA · SEC/CFTC joint guidance (March 2026) · CLARITY Act · KSA/UAE

Core frameworks:   reasoning patterns · autonomous agent framework · modifier system
                   seven-step reasoning chain · pre-match signal framework · athlete framework
Platform:          MCP server tools · data connectors · API providers · Chiliz Agent Kit
                   social intelligence · web agent connectors · fraud signals
Community:         130 calibration records (96%+ accuracy, 21 sports) · benchmarks
Developer tools:   application blueprints · agentic workflow patterns · agent prompts
                   copy-paste templates · compressed summaries · Skills API

{
  "direction":           "HOME",
  "adjusted_score":      72.4,
  "sms":                 79,
  "recommended_action":  "ENTER",
  "composite_modifier":  1.10,
  "confidence_level":    "HIGH",
  "signal_class":        "EXECUTION",

  "modifiers_applied": {
    "athlete_modifier":     1.10,
    "macro_modifier":       1.00,
    "venue_modifier":       1.05,
    "officiating_modifier": 1.02
  },

  "flags": {
    "lineup_unconfirmed":    false,
    "ftp_path2_active":      true,
    "supply_event_type":     "REDUCTION",
    "macro_override_active": false
  }
}

Full schema: core/confidence-output-schema.md

SportMind exposes its intelligence as MCP tools for Claude Desktop and any MCP-compatible AI agent.

→ Standalone MCP server repository: github.com/SportMind/mcp-server

Confirmed working: 2026 UCL Final — all five layers loaded, SMS 100, pre-match signal produced 48 hours before kickoff. Direction CORRECT.

The server is also available embedded in the core library:

git clone https://github.com/SportMind/SportMind
pip install mcp aiohttp

# stdio (Claude Desktop / Claude Code)
python scripts/sportmind_mcp.py

# HTTP/SSE (hosted agents)
python scripts/sportmind_mcp.py --http --port 3001

Tools: sportmind_pre_match · sportmind_signal · sportmind_macro · sportmind_fan_token_lookup · sportmind_sentiment_snapshot · and more.

→ Standalone repo with full setup guide: github.com/SportMind/mcp-server → Embedded deployment guide: MCP-SERVER.md

Data connections: platform/data-connector-templates.md — lineup data, fan token TVL, macro state.

Execution layer: platform/chiliz-agent-kit-integration.md — SportMind intelligence → Chiliz Agent Kit → on-chain execution.

Web agents: platform/web-agent-connectors.md — lineup confirmation (T-2h), PATH_2 supply verification, exchange monitoring, regulatory/macro monitoring.

MCP deployment: platform/sportmind-mcp-deployment.md — live endpoint in 30 minutes.

Compatible with: Claude · GPT-4 · Gemini · LangChain · CrewAI · AutoGen · any LLM (skills are structured markdown, not API wrappers).

130 records. 96%+ accuracy. Zero wrong-direction records outside European football draws.

All records are in community/calibration-data/ — publicly verifiable, pre-submitted before real matches. Includes all wrong predictions with root-cause analysis. Not cherry-picked.

The most recent verified record: UCL Final 2026 — PSG vs Arsenal. SportMind predicted PSG. PSG won. Record #130 verified. See the full signal and outcome: sportmind.dev/first-record/

Eight modifiers with zero wrong-direction records across their entire evidence base: qualifying_delta (F1) · india_pakistan ×2.00 · morning_skate (NHL) · dew_factor (cricket) · taper_modifier (swimming) · raider_primacy (kabaddi) · goalkeeper_save_rate (handball) · superspeedway_specialist (NASCAR)

The fastest contribution: one calibration record. No coding required. See FIRST-RECORD-CHALLENGE.md.

Found a framework gap? See CONTRIBUTING-GAPS.md for how to report missing intelligence frameworks.

What the community needs most:

• Football calibration records (athlete_modifier: 25/50 threshold)
• Cricket dew_factor records (evening T20 matches)
• Records from any underrepresented sport (rowing, netball, kabaddi)

Full process: community/calibration-data/CONTRIBUTING.md Recognition: community/CONTRIBUTORS.md

MIT — free to use, modify, and redistribute for any purpose.

SportMind does not collect or process personal data.

Some SportMind frameworks reason about holder behaviour and on-chain wallet activity. Developers building applications that combine these frameworks with identifying information may be processing personal data under GDPR, UK GDPR, CCPA, or applicable local law.

See core/privacy-notice.md for a full developer privacy reference.

SportMind is MIT licensed. Developers are responsible for their own compliance with data protection law when building on this library.

WHO-USES-THIS.md → FIRST-RECORD-CHALLENGE.md → examples/starter-pack/

SportMind is an independent open-source project. Not affiliated with Chiliz, Socios, or any sports data provider, though designed to complement them.