ACE-Step MCP server has 4 tools with mixed quality. generate_music has detailed parameter schemas with descriptions, defaults, and ranges, strong foundation. However, format_lyrics, create_random_sample, and list_models have minimal or missing parameter documentation. Tool descriptions are present but generic (10-50 chars for some). No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Naming is clear (verb_noun pattern), but descriptions lack LLM-optimized context about when/why to use each tool. Per-tool average: generate_music=68, format_lyrics=42, create_random_sample=38, list_models=45.

• Add output schema documentation for all tools. Example for generate_music: 'Returns {"paths": [string], "duration_seconds": float, "format": string, "seed_used": int}' so LLM knows what fields to expect.
• Expand format_lyrics and create_random_sample parameter descriptions. For use_cot: 'Enable chain-of-thought reasoning to improve lyric coherence and structure (slower but higher quality).' For sample_mode: 'simple_mode generates random parameters; custom_mode allows fine-tuning of all parameters.'
• Add destructiveHint annotation to generate_music and document output_path behavior. Clarify: 'Files are written to disk and cannot be undone. Specify output_path to control location.'
• Document list_models output schema: 'Returns array of {"name": string, "description": string, "parameters": object}' so agent knows how to interpret available models.
• Add error recovery hints to generate_music description: 'If generation fails, check available models with list_models(). If disk space error, specify a different output_path.'