MCP Ollama Consult Server

# MCP Toolkit Enhancement Utilities This directory contains utilities for AI-assisted toolkit enhancement planning. ## Scripts ### `ask_kimi.js` Single-stage consultation with kimi-k2-thinking:cloud for strategic tool suggestions. **Output:** Markdown-formatted suggestions optimized for AI parsing ```bash # Preview the prompt npm run ask-kimi # Execute consultation npm run ask-kimi:run ``` ### `ask_kimi_then_qwen.js` ⭐ **Two-stage AI collaboration workflow:** 1. **Stage 1 (Kimi - Thinking):** Analyzes toolkit gaps and suggests 5-7 new tools in structured JSON format 2. **Stage 2 (Qwen - Instruction):** Takes the suggestions and generates detailed implementation instructions **Outputs:** - `.tmp/toolkit_enhancement_plan.json` - Full analysis + instructions - `TOOLKIT_ENHANCEMENT_PLAN.md` - Human-readable markdown report ```bash npm run ask-kimi-then-qwen ``` **What it does:** - Consults kimi with a JSON-structured prompt requesting tool suggestions - Parses kimi's response and extracts structured tool specifications - Feeds those specs to qwen3-vl:235b-instruct-cloud for implementation guidance - Generates copy-paste ready TypeScript code and step-by-step instructions - Creates both JSON and Markdown outputs for different use cases **Expected duration:** 4-5 minutes total (2-3 min kimi + 1-2 min qwen) ## Workflow Design The two-stage approach leverages model specialization: - **Kimi (k2-thinking):** Deep reasoning for strategic analysis - What tools are missing? - What would provide the most value? - How should implementation be prioritized? - **Qwen (3-vl:235b-instruct):** Instruction-following for tactical implementation - How do we build this? - What's the file structure? - What code needs to be written? ## Output Format ### Kimi Output (JSON) ```json { "analysis_summary": "Overview of gaps", "suggested_tools": [ { "tool_name": "consult_chain_ollama", "display_name": "Sequential Consultation Chain", "category": "collaboration", "priority": "P1", "description": "...", "parameters": { ... }, "use_cases": [ ... ], "implementation": { ... } } ], "implementation_roadmap": { "phase_1": ["tool_1", "tool_2"], "phase_2": ["tool_3"] } } ``` ### Qwen Output (Markdown) ```markdown # Implementation Guide: [Tool Name] ## 1. File Structure ## 2. Handler Implementation ## 3. Service Integration ## 4. Tool Registration ## 5. Type Definitions ## 6. Testing Strategy ## 7. Implementation Order ``` ## Tips - Run `npm run build` before executing these scripts - Ensure Ollama is running and models are available - Check `.tmp/` directory for intermediate results - The markdown report is great for human review - The JSON output is perfect for feeding into other automation ## Extending To modify the prompts: - Edit `buildPrompt()` in `ask_kimi.js` for strategic analysis - Edit `buildQwenPrompt()` in `ask_kimi_then_qwen.js` for implementation instructions To add more stages: - Follow the pattern in `ask_kimi_then_qwen.js` - Use `consultation_type: 'thinking'` or `'instruction'` as appropriate - Chain results through JSON parsing and formatting