The pyLLMSearch server has 4 tools with basic functionality but significant definition quality gaps. All tools have names starting with verbs (rag_*) and descriptions are present. However, descriptions are generic and lack actionable guidance for LLM selection. Input schemas are minimal, most parameters are undescribed (missing parameter descriptions). Output schemas are not documented. Three tools are nearly identical in naming and purpose (rag_retrieve_chunks, rag_generate_answer, rag_generate_answer_simple), creating ambiguity for LLM tool selection. The tool names use prefixes that don't clearly distinguish their roles from each other. No error handling guidance, no parameter constraints, and no indication of prerequisites or when to use one tool vs another.

• Rename tools to clearly disambiguate: 'search_document_chunks' (raw semantic search), 'generate_qa_answer' (LLM-synthesized answer), and 'generate_qa_answer_fast' (if speed vs quality trade-off). Add 1-2 sentence descriptions explaining the difference: when to use retrieval vs synthesis, performance implications.
• Add description to 'label' parameter explaining: what labels are available (enum or discovery endpoint), what filtering behavior is applied, what happens if no documents match the label.
• Document output schema for all tools: rag_retrieve_chunks should specify { chunks: [{text: string, source: string, score: number}], total_count: number }; rag_generate_answer should return { answer: string, sources: string[], confidence: number }. Include field descriptions.
• Add error handling guidance to all tool descriptions: 'If no documents match, returns empty result with explanation. Use a broader query or check available labels via [discovery tool].' For rag_update_index: 'This operation may take minutes. Call with dry_run=true first to validate.'
• For rag_update_index, add parameters: dry_run (boolean, defaults false) and timeout_seconds (integer, 1-3600, default 300). Add description explaining prerequisites and expected side effects.
• Split rag_generate_answer and rag_generate_answer_simple into one tool with a 'response_length' parameter (enum: brief, detailed) or 'model_size' parameter (enum: fast, accurate) to reduce duplication and let LLM choose trade-off explicitly.
• Add parameter constraints via description: 'question: string (1-500 characters, natural language query)' and 'label: string (alphanumeric, optional, matches document tags)'.