cognitive-tools-mcp / gikendaasowin-aabajichiganan

# /// script # dependencies = [ # "smolagents[mcp,toolkit,litellm]", # "python-dotenv", # ] # /// """ smolagents deliberation demo using the gikendaasowin-aabajichiganan MCP server. This script demonstrates how to integrate smolagents with our custom MCP server that provides advanced deliberation capabilities using various cognitive frameworks. The MCP server offers a "deliberate" tool that implements: - OOReDAct cognitive cycle (Observe-Orient-Reason-Decide-Act) - Multiple prompting strategies (Cache-Augmented Reasoning, Self-Consistency, PAL, etc.) - Scientific Investigation methodologies - Critical Thinking frameworks - Dynamic strategy evaluation system Updated to work with local Ollama models (granite3.3:2b) for privacy and local execution. """ import os from dotenv import load_dotenv from smolagents import MCPClient, CodeAgent, LiteLLMModel from mcp import StdioServerParameters def get_available_ollama_model(): """Get the best available Ollama model from our preferred list.""" import subprocess # Preferred models in order of preference preferred_models = [ "granite3.3:2b", "llama3.2:1b", "granite3.3:latest", "deepseek-r1:latest", "qwen3:latest" ] try: # Get list of available models result = subprocess.run(['ollama', 'list'], capture_output=True, text=True, check=True) available_models = [line.split()[0] for line in result.stdout.strip().split('\n')[1:]] # Find the first preferred model that's available for model in preferred_models: if model in available_models: print(f"🎯 Selected Ollama model: {model}") return model # Fallback to first available model if available_models: model = available_models[0] print(f"⚠️ Using fallback model: {model}") return model raise Exception("No Ollama models found") except subprocess.CalledProcessError: raise Exception("Failed to get Ollama model list. Is Ollama installed and running?") except Exception as e: raise Exception(f"Error checking Ollama models: {e}") def main(): """Main function to demonstrate smolagents with gikendaasowin-aabajichiganan MCP integration.""" # Load environment variables load_dotenv() print("🧠 Smolagents + Gikendaasowin-Aabajichiganan MCP Demo") print("🦬 Using Local Ollama Models for Privacy") print("=" * 60) # Get best available Ollama model try: ollama_model = get_available_ollama_model() except Exception as e: print(f"❌ Ollama setup error: {e}") print("\nSetup instructions:") print("1. Install Ollama: https://ollama.com/") print("2. Pull a model: ollama pull granite3.3:2b") print("3. Start Ollama: ollama serve") return 1 # Configure MCP server parameters for our deliberation server # Using local build instead of npm package for development mcp_server_path = os.path.join(os.path.dirname(os.path.dirname(__file__)), "build", "index.js") server_parameters = StdioServerParameters( command="node", args=[mcp_server_path], env={**os.environ} # Pass through all environment variables ) print("📡 Connecting to Gikendaasowin-Aabajichiganan MCP server...") try: # Use MCPClient with context manager for automatic connection management with MCPClient(server_parameters) as mcp_tools: print(f"✅ Connected! Available tools: {[tool.name for tool in mcp_tools]}") # Initialize the inference model using Ollama with selected model # LiteLLMModel allows us to use local Ollama models print(f"🧠 Initializing Ollama model ({ollama_model})...") model = LiteLLMModel( model_id=f"ollama_chat/{ollama_model}", api_base="http://localhost:11434", # Default Ollama endpoint api_key="ollama" # Required by LiteLLM for Ollama ) # Create CodeAgent with MCP tools and base tools agent = CodeAgent( tools=mcp_tools, # MCP tools from our server model=model, add_base_tools=True, # Include Python interpreter, web search, etc. max_steps=6, # Allow multiple reasoning steps verbosity_level=2, # Show detailed reasoning additional_authorized_imports=["json"] # Allow JSON parsing for MCP results ) print("\n🤖 Agent initialized with deliberation capabilities!") print("Available capabilities:") for tool in agent.tools.values(): print(f" - {tool.name}: {getattr(tool, 'description', 'No description available')}") # Show proper usage of the deliberate tool print("\n📚 Deliberate tool usage:") print(" - Correct: deliberate(input='Your question here')") print(" - With context: deliberate(input='Your question', context='Additional context')") print(" - Note: Returns detailed analysis using OOReDAct cognitive framework") # Demo queries showcasing different aspects of the deliberation tool # Optimized for local models - shorter, more focused queries demo_queries = [ { "title": "🔬 Problem Analysis with Deliberation", "query": "Use the deliberation tool to analyze this problem: I need to choose between Python and JavaScript for a web scraping project. Consider factors like performance, ease of use, and library availability." }, { "title": "🧮 Mathematical Reasoning", "query": "Please use deliberation to solve this step by step: If I save $500 per month at 5% annual interest compounded monthly, how much will I have after 3 years?" }, { "title": "🎯 Decision Framework", "query": "Use the deliberation tool to create a decision framework for choosing a laptop: budget $1000-1500, uses: coding, light gaming, portability important." } ] # Run demo queries for i, demo in enumerate(demo_queries, 1): print(f"\n{demo['title']}") print("-" * 50) print(f"Query: {demo['query']}") print("\n🤔 Agent deliberating...") try: result = agent.run(demo['query']) print("\n✅ Deliberation completed!") print(f"Result: {result}") if i < len(demo_queries): input("\nPress Enter to continue to next demo...") except Exception as e: print(f"❌ Error during deliberation: {e}") continue # Interactive mode print("\n🔄 Interactive Mode") print("-" * 30) print("You can now ask the agent any question. It will use the deliberation tool") print("to provide thoughtful, systematic analysis. Type 'quit' to exit.") while True: try: user_query = input("\n🎯 Your question: ").strip() if user_query.lower() in ['quit', 'exit', 'q']: print("👋 Thanks for using the deliberation demo!") break if not user_query: continue print("\n🤔 Agent deliberating...") result = agent.run(user_query) print(f"\n✅ Response: {result}") except KeyboardInterrupt: print("\n👋 Demo interrupted. Goodbye!") break except Exception as e: print(f"❌ Error: {e}") continue except Exception as e: print(f"❌ Failed to connect to MCP server: {e}") print("\nTroubleshooting tips:") print("1. Ensure Node.js is installed (node --version)") print("2. Check if the MCP package is available: npx @nbiish/gikendaasowin-aabajichiganan-mcp --version") print("3. Verify network connectivity for npx package downloads") return 1 return 0 if __name__ == "__main__": exit(main())