A Model Context Protocol (MCP) server that implements the MAKER methodology for complex task execution using consensus and voting.

MAKER-Council is a pure MCP server that provides advanced decision-making tools to your AI assistant. It implements the concepts from the "MAKER: Massively Decomposed Agentic Processes" paper, allowing for high-quality, consensus-driven responses through voting and judging mechanisms.

• MAKER Methodology Tools: Provides specialized tools like consult_council and solve_with_voting to leverage multiple internal micro-agents for better accuracy.
• Task Decomposition: Includes decompose_task to break down complex objectives into manageable steps.
• Unified Query Interface: Exposes a smart query tool that automatically routes requests to the appropriate internal strategy.
• Pure MCP Implementation: communicating entirely over stdio, making it compatible with any MCP client (Claude Desktop, Cline, etc.).

Clone the repository and install dependencies:

git clone https://github.com/your-repo/maker-council.git
cd maker-council
npm install
npm run build

Create a .env file in the root directory. You must provide an API key for the LLM provider (OpenAI by default).

cp .env.example .env

To use MAKER-Council with your favorite MCP client, add it to your configuration file.

Edit %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "maker-council": {
      "command": "node",
      "args": [
        "path/to/maker-council/dist/index.js"
      ],
      "env": {
        "MAKER_API_KEY": "sk-..."
      }
    }
  }
}

Add the server in the MCP Servers tab:

• Name: maker-council
• Command: node
• Args: path/to/maker-council/dist/index.js
• Env: Add your MAKER_API_KEY

The server includes a real-time web dashboard to monitor requests, voting consensus, and agent performance.

To start the dashboard server:

npm run dashboard

Then open http://localhost:3000 in your browser.

• Real-time Stats: Track total requests, error rates, and latency.
• Request Traces: See step-by-step execution of MAKER algorithms (Voters, Judge, etc.).
• Deep Debugging: Inspect full JSON logs, prompts, and tool inputs/outputs.

Note: The dashboard connects to the local SQLite database. Ensure the main MCP server is running or has been run to generate data.

This server exposes the following tools:

The unified entry point. It automatically routes your prompt to the most appropriate internal strategy based on intent or analysis.

Arguments:

• prompt (string): The main query or task.
• context (object, optional): Additional context (code, history, etc.).
• intent (string, optional): Explicit intent ('decision', 'validation', 'decomposition').
• config (object, optional): Overrides for voters or 'k' value.

Uses the full MAKER algorithm. Multiple micro-agents (voters) generate proposals, and a senior judge synthesizes the best answer using a voting mechanism. Best for complex decisions or architectural questions.

Arguments:

• query (string): The question to be analyzed.
• num_voters (number): Number of microagents (1-10).
• k (number): Voting margin.

Solves a question using ONLY the "First-to-Ahead-by-k" voting mechanism (no judge synthesis). Faster and ideal for objective questions where statistical consensus is sufficient.

Arguments:

• query (string): The question to be solved.
• k (number): Voting margin.

Breaks down complex tasks into atomic, verifiable steps (MAD - Maximal Agentic Decomposition).

Arguments:

• task (string): The task to be decomposed.

maker-council/
├── 📄 .env                  # Environment configuration
├── 📁 src/
│   ├── 📄 index.ts          # Main MCP server entry point
│   ├── 📄 tools.ts          # Tool definitions
│   ├── 📄 logic.ts          # Core MAKER logic implementation
│   ├── 📄 config.ts         # Configuration loader
│   └── 📄 types.ts          # Type definitions
├── 📄 package.json
└── 📄 README.md

This project is an implementation inspired by the concepts in "MAKER: Massively Decomposed Agentic Processes" (arXiv:2511.09030v1) by Meyerson et al., 2025.