Scoring was not performed

• Create explicit tool registration with proper JSON Schema. Example: {'name': 'get_word_vector', 'description': '...', 'inputSchema': {'type': 'object', 'properties': {'word': {'type': 'string', 'description': '...'}, 'model': {'type': 'string', 'enum': ['embeddinggemma:latest', ...]}}, 'required': ['word']}}
• Document output schema for each tool. For get_word_vector, specify: {'type': 'object', 'properties': {'vector': {'type': 'array', 'items': {'type': 'number'}, 'description': 'Embedding vector for the input word'}, 'dimensions': {'type': 'integer'}, 'model_used': {'type': 'string'}}}
• Add enum constraints for 'voice' parameter in kokoro_tts. Replace generic description with: 'Voice identifier (one of: af_heart, am_adam, bf_alice, bm_george, ...)' or use JSON Schema enum field.
• Specify range constraints for numeric parameters. For 'speed' in kokoro_tts, add: 'Speech speed multiplier (0.5 - 2.0, where 1.0 is normal speed)'.
• Add error recovery guidance to descriptions. E.g., 'If word not found, try searching Wikipedia or a thesaurus first.' This guides LLM fallback strategies.
• Implement MCP server scaffolding with proper transport (HTTP/SSE). Use a framework like mcp-server-python or equivalent to expose these tools as a real MCP server, not standalone scripts.
• Add parameter validation and return specific error messages. Example: 'Invalid language: got "fr-fr", must be ISO 639-1 code like "en-us", "es-mx", etc.'
• Expand tool descriptions to 100 - 200 characters explaining WHEN to use each tool. E.g., 'Use to retrieve pre-trained word embeddings for semantic similarity analysis, clustering, or neural network input layers. Requires Ollama service running locally.'