Skip to content

mcp-server.md

Latest commit

 

History

History
138 lines (100 loc) · 3.34 KB

mcp-server.md

File metadata and controls

138 lines (100 loc) · 3.34 KB

MCP Server

duh can run as an MCP (Model Context Protocol) server, allowing AI agents to use consensus-driven decision making as a tool.

Starting the server

duh mcp

This starts the MCP server on stdio, ready for integration with any MCP-compatible client.

Available tools

duh_ask

Run multi-model consensus on a question.

Input:

Parameter Type Default Description
question string required The question to get consensus on
protocol string "consensus" Protocol: consensus, voting, or auto
rounds integer 3 Max consensus rounds

Output: JSON with decision, confidence, dissent, and cost.

Example call:

{
  "name": "duh_ask",
  "arguments": {
    "question": "What is the best authentication strategy for a REST API?",
    "protocol": "consensus",
    "rounds": 2
  }
}

Example response:

{
  "decision": "Use short-lived JWTs with refresh token rotation...",
  "confidence": 0.85,
  "dissent": "API keys may be simpler for server-to-server communication",
  "cost": 0.0342
}

duh_recall

Search past consensus decisions by keyword.

Input:

Parameter Type Default Description
query string required Search query
limit integer 10 Max results

Output: JSON array of matching threads with thread_id, question, status, and optionally decision and confidence.

duh_threads

List past consensus threads.

Input:

Parameter Type Default Description
status string all Filter: active, complete, or failed
limit integer 20 Max results

Output: JSON array of threads with thread_id, question, status, and created_at.

Configuration for Claude Desktop

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json on macOS):

{
  "mcpServers": {
    "duh": {
      "command": "duh",
      "args": ["mcp"]
    }
  }
}

If duh is installed in a virtual environment, use the full path:

{
  "mcpServers": {
    "duh": {
      "command": "/path/to/venv/bin/duh",
      "args": ["mcp"]
    }
  }
}

Configuration for Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "duh": {
      "command": "duh",
      "args": ["mcp"]
    }
  }
}

Use cases

  • Architecture decisions -- Have your AI agent consult multiple models before recommending a database, framework, or design pattern.
  • Code review consensus -- Get multi-model agreement on whether a code change is safe to merge.
  • Research synthesis -- Ask complex questions and get a debated, revised answer rather than a single-model response.
  • Decision recall -- Let your agent search past consensus decisions for relevant context.

How it works

When an AI agent calls duh_ask, duh:

  1. Loads your ~/.config/duh/config.toml configuration
  2. Registers all configured providers (Anthropic, OpenAI, Google, Mistral, local models)
  3. Runs the consensus protocol (PROPOSE, CHALLENGE, REVISE, COMMIT)
  4. Returns the consensus decision as structured JSON

The agent receives a well-debated answer rather than a single model's opinion.