genpark-multi-armed-bandit-ucb1-thompson-skill

Multi-armed bandit decision engine with Upper Confidence Bound 1 (UCB1) and Bayesian Thompson Sampling.

Part of the GenPark AI Agent Skills Matrix. Production-ready, zero external dependencies, native Python 3.9+ standard library.

Architecture

flowchart TD
    A[Environment State / Reward] --> B[RL Policy & Value Estimator]
    B --> C[Bellman / UCT Tree Search Kernel]
    C --> D[Optimal Action Selection]
    D --> E[MCP Protocol Endpoint]

Features

• Zero Third-Party Dependencies: Pure Python standard library (math, random).
• Optimal Decision Making: Value iteration convergence, Q-learning TD updates, and UCT search.
• Native MCP Protocol Support: Integrated JSON-RPC 2.0 stdio server ready for Claude Desktop, Cursor, and Windsurf.

Installation

pip install genpark-multi-armed-bandit-ucb1-thompson-skill

Or clone directly:

git clone https://github.com/alphaparkinc/genpark-multi-armed-bandit-ucb1-thompson-skill.git
cd genpark-multi-armed-bandit-ucb1-thompson-skill
python example_usage.py

Quick Start

from client import *
# Refer to example_usage.py for end-to-end execution

Model Context Protocol (MCP) Setup

Add to your claude_desktop_config.json or cursor.json:

{
  "mcpServers": {
    "genpark-multi-armed-bandit-ucb1-thompson-skill": {
      "command": "python",
      "args": ["-m", "genpark-multi-armed-bandit-ucb1-thompson-skill.mcp_server"]
    }
  }
}

License

MIT License. Copyright (c) 2026 AlphaPark Inc. & Alpha-Park.