MCP Ollama Consult Server

Instructions

Implementation Reference

• src/handlers/ConsultOllamaHandler.ts:11-110 (handler)

  Core handler class implementing the execution logic for 'consult_ollama' tool: validates inputs, auto-selects models based on consultation_type, checks model availability with fallbacks, calls OllamaService.consult(), and formats MCP response.
  export class ConsultOllamaHandler extends BaseHandler { private modelValidator: ModelValidator; constructor( private ollamaService: OllamaService, modelValidator?: ModelValidator ) { super(); this.modelValidator = modelValidator || new ModelValidator(this.ollamaService.getConfig()); } async handle(params: unknown): Promise<MCPResponse> { // Validate and cast parameters const typedParams = params as Record<string, unknown>; try { // Validate required parameters this.validateRequired(typedParams, ['prompt']); let model = (typedParams.model as string) || ''; const consultationType = typedParams.consultation_type as string | undefined; // Auto-select model based on consultation type if (consultationType === 'thinking') { model = 'kimi-k2-thinking:cloud'; } else if (consultationType === 'instruction') { model = 'qwen3-vl:235b-instruct-cloud'; } else if (!model) { // If no model specified and no consultation type, use default model = await this.modelValidator.getDefaultModel(); } // Validate model availability if (model) { const isAvailable = await this.modelValidator.isModelAvailable(model); if (!isAvailable) { const suggestions = await this.modelValidator.getSuggestions(3); const typeInfo = consultationType ? ` (auto-selected for ${consultationType} consultation)` : ''; return { content: [ { type: 'text', text: `Model '${model}'${typeInfo} is not available. Available models: ${suggestions.join(', ')}. Please install the model or choose an available one.`, }, ], isError: true, }; } } // Build consult request const request: ConsultRequest = { model, prompt: typedParams.prompt as string, stream: false, }; // Include system prompt if provided if (typedParams.system_prompt) { request.systemPrompt = typedParams.system_prompt as string; } // Include temperature if provided if (typeof typedParams.temperature === 'number') { request.temperature = typedParams.temperature; } // Include timeout if provided if (typeof typedParams.timeout_ms === 'number') { request.timeout = typedParams.timeout_ms; } // Call Ollama service const response = await this.ollamaService.consult(request); // Return formatted response return { content: [ { type: 'text', text: response.response, }, ], }; } catch (error) { // Handle errors with consistent formatting const message = error instanceof Error ? error.message : 'Unknown error'; return { content: [ { type: 'text', text: `Error consulting Ollama: ${message}`, }, ], isError: true, }; } }
• src/handlers/callToolHandler.ts:34-37 (registration)

  Tool registration/dispatch within CallToolHandler.handle(): instantiates ConsultOllamaHandler and delegates the call for 'consult_ollama'.
  case 'consult_ollama': { const handler = new ConsultOllamaHandler(this.ollamaService, this.modelValidator); return await handler.handle(args); }
• src/handlers/listToolsHandler.ts:8-69 (schema)

  Input schema and description for 'consult_ollama' tool as returned by ListToolsHandler, defining parameters like consultation_type, model, prompt, system_prompt, context, temperature, timeout_ms.
  name: 'consult_ollama', description: 'Consult with Ollama AI models for architectural decisions, code reviews, and design discussions. Supports sequential chaining of consultations for complex multi-step reasoning.', inputSchema: { type: 'object', properties: { consultation_type: { type: 'string', enum: ['thinking', 'instruction', 'general'], description: 'Type of consultation: "thinking" (uses kimi-k2-thinking:cloud for reasoning tasks), "instruction" (uses qwen3-vl:235b-instruct-cloud for instruction-following), or "general" (uses specified model or default). If specified, overrides the model parameter.', }, model: { type: 'string', description: 'Model to use (e.g., "qwen2.5-coder:7b-cloud"). If not specified and no consultation_type, uses the first available model. Must be a cloud model (ends with :cloud or -cloud) or locally installed.', }, prompt: { type: 'string', description: 'Your question or prompt for the AI model. Can reference previous consultation results.', }, system_prompt: { type: 'string', description: 'Optional system prompt to guide model behavior', }, context: { type: 'object', description: 'Optional context including code, previous results, and metadata', properties: { code: { type: 'string', description: 'Code snippet or file content for analysis', }, language: { type: 'string', description: 'Programming language of the code', }, previous_results: { type: 'array', description: 'Results from previous consultations in the chain', items: { type: 'string' }, }, metadata: { type: 'object', description: 'Additional contextual information', }, }, }, temperature: { type: 'number', description: 'Sampling temperature (0.0-2.0, default: 0.7)', }, timeout_ms: { type: 'number', description: 'Request timeout in milliseconds (default: 60000). Increase for complex prompts with system prompts (e.g., 120000-300000 for complex reasoning)', }, }, required: ['prompt'], }, },