Server has 3 unique tools (save_memory, recall_memory, search_docs) listed 6 times due to duplication in the tool list. Descriptions exist but are problematic: they contain Chinese text with 'CRITICAL MUST' directives that attempt to force LLM behavior rather than document tool purpose. Descriptions are 150-200 chars but mix imperative instructions ('必须优先调用') with functional documentation. Input schemas are present with basic type info (string parameters) but lack enums, ranges, or validation constraints. No output schemas documented. No error handling guidance. No tool annotations. Naming is acceptable (verb_noun pattern) but descriptions violate the pattern:tool-description principle by prescribing LLM behavior instead of stating what the tool does.

• Rewrite descriptions to document tool purpose, not LLM behavior. Example: 'save_memory: Persist user preferences, code habits, or context facts to long-term memory for retrieval in future sessions. Call when the user explicitly asks to remember something or shares personal context.'
• Add output schemas documenting returned fields. Example for recall_memory: {type: 'object', properties: {memories: {type: 'array', items: {type: 'string'}}, count: {type: 'integer'}}}
• Add validation constraints to input parameters. Example: 'query' should have minLength: 1, maxLength: 256; 'fact' should have minLength: 5, maxLength: 2048.
• Implement error handling that categorizes failures. Return structured errors: {error: 'not_found', message: 'No memories match query. Try a broader keyword.', retryable: false}
• Remove duplicate tool registrations. Verify that save_memory, recall_memory, and search_docs are registered exactly once in the FastMCP server.