This RAE MCP server has three tools for Spanish dictionary lookups. Tool naming is clear and action-oriented (get_word_info, get_daily_word, get_random_word), following verb_noun convention. However, the server has critical gaps in schema completeness, parameter descriptions, and output documentation. Descriptions are present but generic. Most critically, parameter descriptions are missing or minimal, the 'min_length' and 'max_length' parameters in get_random_word lack descriptions in the visible schema. The server logs tool calls via zerolog but provides no structured error recovery guidance. Input validation exists (type checking, nil guards) but error messages are generic. Output formatting is present (formatWordInfo, formatDailyWord) but the output schemas themselves are not documented in tool definitions, only string JSON dumps are returned, making it impossible for LLMs to know what fields to expect. The codebase shows reasonable engineering (error propagation, logging), but from an MCP tool definition standpoint, the definition quality is below production baseline.

• Add explicit parameter descriptions to all three tools' input schemas. For get_random_word: 'min_length: Minimum word length in characters (1 - 50, optional, default=1). max_length: Maximum word length in characters (1 - 100, optional, default=100).'
• Document the output schema for each tool. Example for get_word_info: 'Returns a JSON object with fields: word (string), definitions (array of {definition, partOfSpeech, examples}), etymology (string), synonyms (array). Stripped of audit metadata and pagination fields.'
• Enhance error messages with recovery guidance. Example: 'Word "xyz" not found. Try get_daily_word() for an example, or search with partial spelling.' Categorize errors as retryable (network) vs. user-fixable (spelling).
• Declare the 'lang' parameter explicitly in get_word_info schema if supported, or remove it from the code. If only Spanish is supported, state 'Spanish only' in the tool description.
• Add result limits and explain pagination. E.g., 'Returns up to 10 definitions per word. If more exist, a results are truncated. Consider filtering by part-of-speech in a future call.'
• Validate numeric parameters early in get_random_word. If min_length or max_length are out of range (e.g., > 1000), return: 'Invalid min_length: got 9999, must be 1 - 100.' This guides the LLM to self-correct.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry