fiction-forge demonstrates solid naming conventions and comprehensive parameter schemas, but suffers from generic descriptions that lack actionable detail for LLM tool selection. All five tools start with action verbs (get_, search_) and have well-formed JSON Schema inputs with type constraints. However, descriptions are formulaic and fail to distinguish between tools effectively, e.g., 'Get character voice notes' doesn't explain WHEN to call this vs. search_bible. Parameter descriptions are present but minimal (often single-phrase), missing examples of expected values, ranges, and dependency relationships. No output schemas are documented. Error handling is entirely absent from visible code. The server is READ_ONLY with no destructive operations, which mitigates risk, but the lack of error guidance and output structure documentation prevents agents from planning multi-step workflows confidently.

• Expand tool descriptions to 150 - 250 chars, following this template: '[ACTION] [WHEN/WHY] [PREREQUISITES] [RETURNS]'. Example: 'search_bible: Search across story bible, characters, worldbuilding, and reference docs to answer questions about the story world. Use this when you need general knowledge about lore, history, or character background. Returns matching sections with source file and heading. If no results match, try broader search terms or call get_character/get_chapter_context for specific character/chapter info.'
• Document the output schema for each tool. Add a 'Returns' section to each description specifying field names, types, and whether results are paginated. E.g., 'search_bible returns: {results: [{heading: string, source: string, text: string, chapter_range?: [int, int]}], total_count: int, search_quality: 'high'|'partial'|'no_match'}'.
• Enhance parameter descriptions with format, range, and dependency notes. E.g., 'query (required, 1 - 100 chars): search terms; supports fuzzy matching; leave empty to return all chapters' and 'chapter (optional, 1 - 999): if provided, returns character state AT that chapter; omit to get latest state'.
• Add error handling guidance to descriptions. E.g., 'If character name is unknown, returns empty result, try search_bible to find character aliases or check spelling.'
• Clarify tool selection cues: When should an agent call search_bible vs get_character? Suggest: 'Use search_bible for general questions (settings, plot, themes, magic systems). Use get_character for character-specific info (voice, arc position, aliases). Use get_chapter_context to read actual prose.'