MCP Ollama Consult Server

# MCP Consult - AI Model Quick Start Guide Welcome! This guide helps AI models understand how to effectively use the mcp-consult server. ## What is mcp-consult? A Model Context Protocol (MCP) server that lets you: - 🤖 Consult with local Ollama AI models - 🔄 Compare responses from multiple models - 💾 Store and retrieve context across sessions - 🧠 Build sophisticated multi-model reasoning workflows ## Available Tools ### 1. `consult_ollama` - Ask Any Model Ask a specific Ollama model a question. **Parameters:** - `prompt` (required): Your question - `model` (optional): Model name (e.g., "llama3.2", "qwen2.5-coder:7b") - `system_prompt` (optional): System instruction to guide the model - `temperature` (optional): Creativity level (0.0 = focused, 2.0 = creative) **Example:** ```json { "tool": "consult_ollama", "arguments": { "model": "llama3.2", "prompt": "Design a scalable API for a real-time chat application", "system_prompt": "You are an expert software architect", "temperature": 0.7 } } ``` **Response Format:** ```json { "content": [ { "type": "text", "text": "The response from the model..." } ] } ``` --- ### 2. `list_ollama_models` - See Available Models Get a list of all models running on the local Ollama instance. **Parameters:** None **Example:** ```json { "tool": "list_ollama_models", "arguments": {} } ``` **Response:** Returns a JSON array of available models with their names and sizes. --- ### 3. `compare_ollama_responses` - Get Multiple Perspectives Ask multiple models the same question and compare their responses. **Parameters:** - `models` (required): Array of model names - `prompt` (required): The question - `context` (optional): Additional context **Example:** ```json { "tool": "compare_ollama_responses", "arguments": { "models": ["llama3.2", "qwen2.5-coder:7b", "deepseek-v3.1"], "prompt": "What are the pros and cons of microservices?", "context": { "company_size": "startup", "team_experience": "intermediate" } } } ``` **Response:** Returns an array of responses, one per model. If a model fails, you'll see which one and an error message. --- ### 4. `remember_context` - Store Data for Later Save context that persists for the rest of the session. **Parameters:** - `key` (required): A unique identifier - `value` (required): The data to store - `metadata` (optional): Extra information about the data **Example:** ```json { "tool": "remember_context", "arguments": { "key": "project_requirements", "value": "Build a REST API with authentication", "metadata": { "priority": "high", "added_at": "2025-11-17" } } } ``` --- ## Error Handling If something goes wrong, the response will include `isError: true`: ```json { "content": [ { "type": "text", "text": "Error: model 'non-existent-model' not found" } ], "isError": true } ``` **Always check for `isError` flag** to handle failures gracefully. --- ## Tips for Best Results ### 1. Use System Prompts Wisely ```json { "system_prompt": "You are an expert in cloud architecture. Be concise but thorough." } ``` ### 2. Adjust Temperature for Your Needs - **0.0-0.5**: Focused, factual answers (good for technical questions) - **0.7**: Balanced (good for general questions) - **1.0-2.0**: Creative, diverse answers (good for brainstorming) ### 3. Compare Models for Complex Questions Use `compare_ollama_responses` when you need: - Multiple perspectives - Cross-validation of answers - Comprehensive analysis ### 4. Build on Stored Context Use `remember_context` to: - Maintain project state across tool calls - Share information between different consultations - Build context chains for complex reasoning --- ## Example Workflow ### Workflow: Design API with Architectural Review ``` 1. Get available models → list_ollama_models 2. Consult architect model → consult_ollama (llama3.2, "Design a REST API for...") 3. Store design in context → remember_context (key: "api_design", value: design_response) 4. Get security review from another model → consult_ollama (qwen2.5-coder, "Review security of this API: {api_design}") 5. Compare approaches from multiple models → compare_ollama_responses (all_models, "Optimization suggestions for...") ``` --- ## Common Questions **Q: How long can I wait for a response?** A: Default timeout is 30 seconds. For complex tasks, the server may need longer. **Q: Can I use variables in prompts?** A: Use `{context_key}` syntax to reference stored context from `remember_context`. **Q: What if a model I want isn't available?** A: Use `list_ollama_models` to see what's available, or ask to install it on the Ollama instance. **Q: How much context can I store?** A: Store as much as needed, but remember it persists only for the current session. **Q: Can responses from one call be used in another?** A: Yes! Store them with `remember_context` and reference them in follow-up prompts. --- ## Performance Notes - First call to a model may be slower (model loading) - Subsequent calls are faster (model stays in memory) - `compare_ollama_responses` runs all requests in parallel for speed - Large models (>30GB) need more time; increase timeout if needed --- ## Getting Help If you encounter issues: 1. Check that Ollama is running (`http://localhost:11434` by default) 2. Verify the model exists with `list_ollama_models` 3. Try a simpler prompt first to isolate issues 4. Check error messages - they include helpful details --- **Happy consulting! 🚀** For more advanced usage, see the main [README.md](README.md).