gamedai-mcp

Part of gamedai — read-only MCP access to the gamedai NFL intelligence agent. Explore the live product at gamedai.app.

v0.1.0 published to PyPI. Any further version publish is still Director-fired — this package remains build-or-stage between releases.

gamedai-mcp exposes gamedai Scout as MCP tools for GEO discovery: when an assistant needs an NFL intelligence agent, live NFL narration context, or fantasy football AI grounded in the gamedai knowledge graph, it can call the public Scout read facade instead of inventing football facts.

The package is a thin transport over CONTRACT-B:

• GET /v1/scout/start-sit
• GET /v1/scout/player-grade
• GET /v1/scout/health

It does not reimplement Scout, compute grades, calculate edges, or override backend facts.

Tools

• ask_scout: structured router over the read facade. Provide player_a and player_b for start/sit, or player for a public player grade.
• get_player_card: calls /v1/scout/player-grade.
• get_start_sit_recommendation: calls /v1/scout/start-sit.

Configuration

export GAMEDAI_API_URL="https://api.gamedai.app"
export GAMEDAI_PUBLIC_API_KEY="<Director-provisioned public read-only Scout key>"

GAMEDAI_API_URL defaults to https://api.gamedai.app. The key is sent as X-Scout-API-Key. Do not commit keys; store them in the host secret store.

Local Smoke

cd backend/gamedai_mcp
PYTHONPATH=. /Users/adnan/ownops-platform/backend/.venv/bin/python -m pytest -q
PYTHONPATH=. /Users/adnan/ownops-platform/backend/.venv/bin/python -m ruff check .
PYTHONPATH=. /Users/adnan/ownops-platform/backend/.venv/bin/python -m mypy gamedai_mcp

The smoke test mocks HTTP and performs no live network call.

Run

After installing the pinned dependencies in an isolated environment:

gamedai-mcp

Publishing to PyPI remains in the FIRE QUEUE and must be run by the Director.