ptcg-meta-bench

name	ptcg-meta-bench
description	Use this repo to evaluate Pokemon TCG AI Battle agents locally through the ptcg CLI or MCP server.

ptcg-meta-bench Agent Skill

Use CLI plus --json for terminal agents. Use MCP only when the host client has MCP support.

Required Checks

ptcg doctor --json
ptcg decks --json

Output Schemas

ptcg decks --json returns:

{
  "status": "ok",
  "deck_count": 10,
  "weight_sum": 81.8,
  "decks": [
    {
      "id": "01_archaludon_duraludon",
      "name": "Archaludon ex / Duraludon",
      "archetype": "archaludon",
      "weight": 15.5,
      "path": "decks/meta/01_archaludon_duraludon.csv"
    }
  ]
}

ptcg demo --json and ptcg bench --json return:

{
  "status": "ok",
  "version": "1.1.0",
  "agent": "examples/agents/simple_baseline",
  "games_per_deck": 2,
  "deck_count": 10,
  "weight_sum": 81.8,
  "meta_weighted_win_rate": 1.0,
  "plain_english_summary": "Your agent beats the meta 100% weighted.",
  "run_dir": "/absolute/path/to/runs/<timestamp>-ptcg_demo",
  "rows": [
    {
      "id": "01_archaludon_duraludon",
      "weight": 15.5,
      "games": 2,
      "completed": 2,
      "wins": 2,
      "losses": 0,
      "draws": 0,
      "errors": 0,
      "engine_errors": 0,
      "max_decisions": 0,
      "win_rate": 1.0
    }
  ]
}

MCP

ptcg mcp

Tool names:

• list_decks
• run_benchmark
• play_game
• get_last_results

Claude Desktop config:

{
  "mcpServers": {
    "ptcg-meta-bench": {
      "command": "ptcg",
      "args": ["mcp"]
    }
  }
}