Homelab MCP Server

name: "Ollama Local LLM Server" description: "Self-hosted large language model server optimized for Raspberry Pi and homelab use" category: "ai" priority: "high" version: "0.1.17" homepage: "https://ollama.ai" # System requirements (Pi 4+ recommended) requirements: ports: [11434] memory_gb: 4 # Minimum for 7B models disk_gb: 20 # For model storage cpu_cores: 4 # Pi 4+ recommended dependencies: - docker # Default configuration default_port: 11434 default_config: models_path: "/opt/ollama/models" max_model_size: "7b" # Pi limitation api_host: "0.0.0.0" # Installation method installation: method: "docker-compose" docker_compose: version: "3.8" services: ollama: container_name: "ollama" image: "ollama/ollama:latest" restart: "unless-stopped" ports: - "11434:11434" environment: - "OLLAMA_HOST=0.0.0.0" - "OLLAMA_MODELS=/models" volumes: - "/opt/ollama/models:/root/.ollama" - "/opt/ollama/config:/etc/ollama" # Pi-specific optimizations deploy: resources: limits: memory: 3G # Leave 1GB for system healthcheck: test: ["CMD", "curl", "-f", "http://localhost:11434/api/tags"] interval: 30s timeout: 10s retries: 3 networks: - ollama-network networks: ollama-network: driver: bridge # Pi-optimized model recommendations recommended_models: small_models: # For Pi 4 (4GB RAM) - name: "tinyllama" size: "637MB" description: "Tiny but capable model" - name: "phi" size: "1.6GB" description: "Microsoft's small language model" - name: "gemma:2b" size: "1.4GB" description: "Google's lightweight model" medium_models: # For Pi 5 (8GB RAM) - name: "llama2:7b" size: "3.8GB" description: "Meta's Llama 2 7B model" - name: "mistral:7b" size: "4.1GB" description: "Mistral 7B instruct model" - name: "codellama:7b" size: "3.8GB" description: "Code generation model" # Post-installation setup post_install: initial_setup: - "Access Ollama API at http://{hostname}:11434" - "Install a model: curl -X POST http://{hostname}:11434/api/pull -d '{\"name\": \"tinyllama\"}'" - "Test generation: curl -X POST http://{hostname}:11434/api/generate -d '{\"model\": \"tinyllama\", \"prompt\": \"Hello!\"}'" - "Web UI available via separate container or Open WebUI" model_management: - "List models: curl http://{hostname}:11434/api/tags" - "Remove model: curl -X DELETE http://{hostname}:11434/api/delete -d '{\"name\": \"model_name\"}'" - "Monitor storage: docker exec ollama ollama list" # Performance optimization for Pi performance_tuning: pi_4_settings: memory_limit: "3G" swap_file: "2G" # Enable swap for larger models cpu_affinity: "0-3" # Use all cores pi_5_settings: memory_limit: "6G" swap_file: "4G" cpu_affinity: "0-3" model_loading: preload_model: "tinyllama" # Load default model on startup context_size: 2048 # Reduce for Pi num_threads: 4 # Match Pi cores # API examples api_examples: generate_text: curl: | curl -X POST http://{hostname}:11434/api/generate \ -H "Content-Type: application/json" \ -d '{ "model": "tinyllama", "prompt": "Explain Raspberry Pi GPIO in simple terms", "stream": false }' chat_completion: curl: | curl -X POST http://{hostname}:11434/api/chat \ -H "Content-Type: application/json" \ -d '{ "model": "tinyllama", "messages": [ {"role": "user", "content": "Help me with a Python script for Pi GPIO"} ] }' # Integration options integrations: open_webui: description: "Web-based chat interface for Ollama" docker_run: | docker run -d \ --name open-webui \ -p 8080:8080 \ -e OLLAMA_API_BASE_URL=http://ollama:11434/api \ --network ollama-network \ ghcr.io/open-webui/open-webui:main langchain_python: description: "Python integration with LangChain" example: | from langchain.llms import Ollama llm = Ollama(base_url="http://{hostname}:11434", model="tinyllama") response = llm("Write a Python function") # Monitoring and maintenance monitoring: api_endpoints: - "/api/tags" # List installed models - "/api/ps" # Show running models - "/api/version" # Ollama version resource_monitoring: - "Memory usage: docker stats ollama" - "Model storage: du -sh /opt/ollama/models" - "API performance: curl -w @curl-format.txt http://{hostname}:11434/api/tags" log_locations: container_logs: "docker logs ollama" model_logs: "/opt/ollama/config/logs" # Troubleshooting troubleshooting: common_issues: - issue: "Out of memory when loading model" solutions: - "Use smaller model (tinyllama, phi)" - "Enable swap: sudo dphys-swapfile swapoff && sudo dphys-swapfile setup && sudo dphys-swapfile swapon" - "Increase container memory limit" - issue: "Model download fails" solutions: - "Check internet connection" - "Verify disk space: df -h" - "Try smaller model first" - "Check Ollama logs: docker logs ollama" - issue: "API not responding" solutions: - "Check container status: docker ps" - "Verify port binding: netstat -tlnp | grep 11434" - "Test local API: curl http://localhost:11434/api/tags" - "Check firewall: sudo ufw status" # Pi-specific considerations pi_considerations: hardware_limits: - "Pi 4 (4GB): Stick to 2B-3B parameter models" - "Pi 5 (8GB): Can handle 7B models with swap" - "Storage: Use fast SD card (Class 10) or USB 3.0 SSD" - "Cooling: Ensure adequate cooling for sustained loads" performance_tips: - "Use GPU acceleration if available (Pi 5)" - "Pre-pull models during off-peak hours" - "Monitor temperature: vcgencmd measure_temp" - "Consider external storage for models" # Security considerations security: network_access: - "Bind to localhost for local-only access" - "Use reverse proxy (nginx) for external access" - "Consider VPN access (Tailscale/WireGuard)" model_safety: - "Only download models from trusted sources" - "Review model cards and capabilities" - "Monitor API usage and logs" - "Consider content filtering for external access"