24 tools with consistent naming (verb_noun pattern: spell_check, morphology, etc.) and adequate descriptions (100-150 chars typical). All tools are READ_ONLY, reducing complexity. However, input schemas are minimal, most accept only a single 'text' or 'word' parameter with basic string type and description. Output schemas are entirely undocumented: no return type specifications, field names, or structure guidance for LLMs. Parameter descriptions are present but generic ('Estonian text to check for X'). Error handling and recovery guidance absent. No tool annotations (readOnlyHint, etc.) despite all being read-only. Composition is sound (each tool does one thing), but lack of output documentation severely limits LLM ability to chain tools or extract structured results.

• Document output schemas for all 24 tools. Specify return type (object, array, string), field names, types, and descriptions. E.g., spell_check should return {errors: [{word: string, suggestions: string[], position: number}], total: number}.
• Add readOnlyHint: true to all tool definitions since all 24 are read-only. This enables client-side caching and optimized retry logic.
• Enhance parameter descriptions with format constraints. E.g., 'Estonian text (UTF-8, max 10,000 chars) to check for spelling errors. Returns list of misspelled words with suggestions.'
• Add error handling guidance to tool descriptions. E.g., 'Returns empty result if text is empty or not valid UTF-8. Timeout after 30s for texts >100KB.'
• For tools returning lists (synonyms, related_words, named_entities, tokenize), document result limits and add optional limit/offset parameters to enable pagination.