🤖 AI Discussion MCP Server

Optimized MCP server for multi-model AI discussions with intelligent context management and token optimization.

✨ Features

• Multi-Model Support: Claude 4.0 (Augment internal), Gemini 2.5 Pro/Flash, DeepSeek R1
• Context Optimization: Intelligent token management and compression
• Cost Efficient: Claude 4.0 free via Augment, optimized pricing for other models
• Parallel Processing: Simultaneous model calls for faster discussions
• Smart Context Sharing: All models receive the same conversation context

🚀 Quick Setup

1. Install Dependencies

pip install -r requirements.txt

2. Configure API Keys

Create or update augment_config.json:

{
  "openrouter_api_key": "your_openrouter_key_for_deepseek_r1",
  "gemini_api_key": "your_google_api_key_for_gemini"
}

3. Add to Augment

Add to your Augment MCP configuration:

{
  "mcpServers": {
    "ai-discussion": {
      "command": "python3",
      "args": ["mcp_discussion_server.py"],
      "cwd": "/path/to/mcp_sdk",
      "env": {
        "OPENROUTER_API_KEY": "your_openrouter_api_key",
        "GEMINI_API_KEY": "your_gemini_api_key"
      }
    }
  }
}

4. Usage in Augment

Thảo luận giữa Claude, Gemini và DeepSeek về cách tối ưu database performance

🛠️ Available Tools

Core Discussion Tools

1. discuss - Multi-Model Discussion

Purpose: Detailed multi-perspective analysis với synthesis Usage: "Hãy thảo luận về [topic] giữa r1 và gemini"

Examples:

• "Thảo luận về microservices vs monolith cho startup"
• "So sánh Python vs Go cho backend API"
• "Discuss pros and cons of TypeScript"

2. chat - Simple Chat

Purpose: Direct conversation với specific model Usage: "Chat với [model] về [topic]"

Examples:

• "Chat với deepseek về async programming"
• "Hỏi r1 về best practices cho REST API"
• "Ask AI about JWT implementation"

3. consensus - Multi-Model Consensus

Purpose: Focused opinion synthesis với clear recommendation Usage: "Lấy consensus về [question]"

Examples:

• "Consensus về database choice cho e-commerce platform"
• "Ý kiến chung về cloud providers (AWS vs GCP vs Azure)"
• "Lấy consensus về programming language cho AI project"

📊 Tool Comparison:

• discuss = Detailed exploration 🔍
• chat = Quick answers ⚡
• consensus = Clear decisions 🎯

API Reference

configure(api_keys: dict)

Configure API keys for the providers.

Parameters:

• api_keys: Dictionary with provider names as keys and API keys as values
  • Supported providers: 'openai', 'gemini', 'openrouter'

chat(prompt: str, model: str, history: list = None)

Start or continue a conversation.

Parameters:

• prompt: The user's message
• model: Model name to use (e.g., 'gpt-4o-mini', 'gemini-1.5-flash')
• history: Previous conversation history (optional)

Returns:

• tuple: (response, new_history) where response is the AI's reply and new_history contains the updated conversation history

get_consensus(prompt: str, models: list[str])

Get consensus from multiple models.

Parameters:

• prompt: The question or topic to get consensus on
• models: List of model names to consult

Returns:

• str: Synthesized consensus response from all models

🎯 Supported Models

Claude (Augment Internal - FREE)

• claude-4 - Claude 4.0 via Augment internal (no API costs)
• claude - General Claude reference

Google Gemini

• gemini-2.5-pro - 2M context window, best quality
• gemini-2.5-flash - 1M context window, fast & cost-effective
• gemini-2.5-flash-preview-04-17 - Preview version
• gemini-2.5-flash-lite-preview-06-17 - Lightweight version

OpenRouter

• deepseek-r1 - Cost-effective reasoning model
• deepseek/deepseek-r1 - Full OpenRouter format

💰 Cost Optimization

Optimized for minimal costs:

• Claude 4.0: $0.00 (free via Augment internal)
• Gemini 2.5 Flash: $0.0008 per 6.5K tokens
• DeepSeek R1: $0.0112 per 6.5K tokens

🚀 Performance Features

• Context Compression: 31% reduction with quality preservation
• Token Caching: Faster estimation with LRU cache
• Parallel Processing: Multiple model calls simultaneously
• Smart Budget Management: Intelligent token allocation

📁 Project Structure

mcp_sdk/
├── core/
│   ├── main_logic.py           # Core chat and consensus functions
│   └── provider_manager.py     # Provider management and routing
├── providers/                  # AI model providers (Google, OpenRouter)
├── system_prompts/             # Optimized system prompts
└── utils/
    ├── token_utils.py          # Model-specific token estimation
    ├── context_compression.py  # Intelligent text compression
    ├── model_optimizer.py      # Per-model optimizations
    ├── token_budget.py         # Budget management
    └── conversation_memory.py  # Context persistence

📚 Documentation

• OPTIMIZATION_SUMMARY.md - Detailed optimization guide and results
• README.md - This file

📄 License

MIT License