Skip to content
Conduit

mcp server

ABBA

Agent-callable sports analytics toolkit -- live data, ensemble predictions, Kelly sizing. pip install, HTTP REST, or MCP server.

paxtonedgar/ABBAcommunitylistedunreviewed
Use tool

Set up

Runs on your computer. Pick your agent and operating system.

Before you start

  • uv (`curl -LsSf https://astral.sh/uv/install.sh | sh`)
~/Library/Application Support/Claude/claude_desktop_config.json
{
  "mcpServers": {
    "abba": {
      "command": "uvx",
      "args": [
        "abba"
      ]
    }
  }
}

same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12

Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs

README

synced Oct 23, 2026

ABBA

ABBA is a DuckDB-backed sports analytics toolkit for agent workflows. It exposes query, prediction, market, and workflow tools through three interfaces:

  • direct Python import
  • a small HTTP server
  • an MCP server

The active runtime path lives under src/abba/server, src/abba/engine, src/abba/storage, and src/abba/workflows.

What It Does

  • Queries games, odds, team stats, rosters, and NHL-specific datasets
  • Produces general game predictions and NHL-specific predictions
  • Calculates EV and Kelly sizing
  • Runs multi-step workflows such as game_prediction, tonights_slate, and betting_strategy
  • Stores local data in DuckDB for repeatable development and testing

Current toolkit surface: 23 callable tools.

Quick Start

Install for local development

git clone https://github.com/paxtonedgar/ABBA.git
cd abba-nhl
pip install -e ".[dev]"

Use from Python

from abba import ABBAToolkit

toolkit = ABBAToolkit(db_path=":memory:", auto_seed=True)

tools = toolkit.list_tools()
games = toolkit.query_games(sport="NHL", status="scheduled")

if games["count"]:
    game_id = games["games"][0]["game_id"]
    prediction = toolkit.nhl_predict_game(game_id)
    odds = toolkit.compare_odds(game_id)

Run the HTTP server

abba-server

Then call:

curl http://localhost:8420/tools
curl -X POST http://localhost:8420/tools/call \
  -H "Content-Type: application/json" \
  -d '{"name":"list_sources","arguments":{}}'

Run the MCP server

abba-mcp

Example MCP config:

{
  "mcpServers": {
    "abba": {
      "command": "abba-mcp",
      "env": {
        "ABBA_DB_PATH": "abba.duckdb"
      }
    }
  }
}

Tool Categories

Data

  • query_games
  • query_odds
  • query_team_stats
  • query_goaltender_stats
  • query_advanced_stats
  • query_cap_data
  • query_roster
  • list_sources
  • describe_dataset
  • refresh_data

Prediction and analysis

  • predict_game
  • nhl_predict_game
  • explain_prediction
  • graph_analysis
  • season_review
  • playoff_odds

Market and session

  • find_value
  • compare_odds
  • calculate_ev
  • kelly_sizing
  • session_budget

Workflows

  • run_workflow
  • list_workflows

Current Scope

The repo is strongest today as:

  • a local agent toolkit
  • an NHL-oriented demo and experimentation surface
  • a testbed for prediction, workflow, and trust-contract checks

The repository also contains experimental and legacy modules outside the main toolkit path. If you are new to the codebase, start with:

  • src/abba/server/toolkit.py
  • src/abba/server/tools/
  • src/abba/engine/
  • src/abba/storage/duckdb.py
  • src/abba/workflows/engine.py

Live Data Notes

The active live-refresh path currently covers:

  • NHL standings and schedule
  • NHL rosters and goalie stats
  • optional odds ingestion when ODDS_API_KEY is set

Development and tests also rely on deterministic seed data.

Development

Run tests

From the repo root:

pytest

For a focused local pass:

pytest tests/test_storage.py tests/test_engine.py tests/test_toolkit.py

Repo layout

src/abba/server/      Public toolkit interfaces
src/abba/engine/      Prediction, EV, Kelly, graph, and NHL logic
src/abba/storage/     DuckDB-backed persistence
src/abba/workflows/   Multi-step workflow orchestration
src/abba/connectors/  Seed and live data ingestion
tests/                Unit, integration, and contract-style checks
docs/                 Project notes, plans, and audit material

Documentation

Start with:

  • docs/README.md
  • docs/architecture-dossier.md

Not every document in docs/ is canonical. Some are historical planning notes or audit artifacts.

License

MIT

↑↓ navigate openesc close