Add quorum-mcp MCP server

[aj-geddes]

Summary

Adds Quorum MCP - a Multi-LLM consensus system for improved AI accuracy and reliability through deliberative synthesis.

Server Details

• Name: quorum-mcp
• Category: ai
• Source: https://github.com/aj-geddes/quorum-mcp
• License: MIT
• Image: mcp/quorum-mcp

Description

Quorum MCP orchestrates multiple AI providers (Claude, GPT-4, Gemini, Ollama, Cohere, Mistral, Novita) through multi-round deliberation to produce consensus-based responses with higher confidence than single-model outputs. This approach is valuable for critical decisions, complex questions requiring multiple perspectives, and validating AI outputs against each other.

Key Features

• Multi-provider consensus: Query 7+ AI providers simultaneously
• Three operational modes:
  • Quick Consensus: Single round parallel queries (fastest)
  • Full Deliberation: Three-round process with cross-review (most thorough)
  • Devil's Advocate: One provider critiques, others respond (explores edge cases)
• Confidence scoring: 0.0-1.0 confidence scores based on consensus strength
• Cost tracking: Per-session USD cost tracking across all providers
• Session management: Retrieve past deliberations by session ID
• Local LLM support: Ollama integration for cost-free experimentation

Tools Provided

1. q_in: Submit a query to the quorum for consensus-based response

   • Parameters: query (required), context (optional), mode (optional: quick_consensus, full_deliberation, devils_advocate)
   • Returns: session_id, status, consensus, confidence score, cost, providers_used

2. q_out: Retrieve consensus results from a quorum session

   • Parameters: session_id (required)
   • Returns: full session details including consensus, confidence, metadata, token usage, cost breakdown

Configuration

Requires at least one API key:

• ANTHROPIC_API_KEY (Claude)
• OPENAI_API_KEY (GPT-4)
• GOOGLE_API_KEY (Gemini)
• COHERE_API_KEY (Cohere)
• MISTRAL_API_KEY (Mistral)
• NOVITA_API_KEY (Novita)
• OLLAMA_ENABLE (default: true, local LLM - no API key required)

Testing

• Tested locally with Docker build
• Verified stdio transport compatibility
• Tools are discoverable without configuration
• Documentation complete and comprehensive
• All validation checks passed

Checklist

• License is MIT or Apache-2.0 (MIT)
• Dockerfile present and builds successfully
• server.yaml follows required schema
• tools.json is complete and valid JSON
• readme.md is comprehensive with usage examples
• Server supports stdio transport (FastMCP default)
• No hardcoded secrets in configuration
• Category is valid (ai)
• Tags are descriptive (multi-llm, consensus, ai, llm-orchestration, deliberation)
• Image name follows mcp/ namespace convention

Use Cases

• Critical decision-making requiring high confidence
• Technical architecture decisions
• Complex problem analysis
• AI output validation
• Code review and security analysis
• Multi-perspective research questions

Performance

• Quick Consensus: ~3-5 seconds, ~$0.01-0.03
• Full Deliberation: ~15-30 seconds, ~$0.05-0.10
• Devil's Advocate: ~10-15 seconds, ~$0.03-0.05

Links

• Source Repository: https://github.com/aj-geddes/quorum-mcp
• Issue Tracker: https://github.com/aj-geddes/quorum-mcp/issues
• MCP Specification: https://modelcontextprotocol.io