The server defines 2 tools with reasonable parameter structure and clear intent, but has significant gaps in parameter descriptions, output schema documentation, and error handling guidance. Both tools are read-only semantic search operations with similar structure. Tool names follow verb_noun convention appropriately (search_messages, search_chat). However, the descriptions lack critical context about when to use one vs. the other, and output schemas are not documented. Parameter descriptions exist but are minimal. No error recovery guidance or validation constraints are evident in the code.

• Document the output schema explicitly. Add a description like: 'Returns an array of message objects, each with fields: document (string, the message text), metadata (object with sender, chat_name, timestamp), distance (float, relevance score 0-1 where lower is more relevant).'
• Replace the 'category' parameter description with an enum constraint listing valid categories: ["personal", "work", "family", ...] or remove the parameter if categories are dynamic.
• Add decision guidance to tool descriptions: 'Use search_messages to search across all your message history. Use search_chat if you already know which conversation to search in.'
• Enhance parameter descriptions with concrete constraints: 'n_results (integer, 1-100, default 10): Number of message chunks to return per query. Higher values may slow down responses.'
• Improve error handling by catching specific exceptions and returning actionable guidance: 'Vector DB connection failed. Ensure the service is running at VECTOR_DB_URL. Error: {details}. Try again in a few seconds.'
• Change 'chat_id' parameter to accept human-readable names: 'chat_id_or_name (string): ID or name of the conversation to search. If you only have a person's name, use search_messages first to find recent conversations.'
• Add validation in call_tool() to enforce n_results bounds: if n_results > 100 or n_results < 1, raise ValueError with guidance.
• Document that distance scores are cosine distances (0-2 range) and show how to interpret relevance. Include examples in the tool description of typical outputs.