Arachne MCP is a STDIO-only proxy to a remote API. While tool schemas are present with types and defaults, descriptions are emoji-heavy marketing copy (10-100 chars) that lack actionable LLM guidance. No parameter descriptions exist, LLMs cannot infer what 'stages', 'kb_id', 'evasion_enabled' mean. Output schemas are undocumented. Error handling is absent from visible code. Tool #26 (arachne_torrent_extract) has NO input schema and a description hinting at copyright violation. Most tools conflate multiple concerns (e.g., arachne_extract does YouTube, audio, video, PDF, DOCX, images in one tool). No tool annotations, no recovery guidance, no pagination hints.

• Add 50-200 character descriptions to every tool explaining WHAT it does, WHEN to use it, and what it returns. Replace emoji-heavy marketing copy with actionable LLM guidance. Example: 'arachne_extract' → 'Intelligently extracts content from any URL (YouTube, audio, video, PDF, DOCX, images). Auto-detects format and returns structured data. Use when you need to ingest diverse media types.'
• Add descriptions to EVERY input parameter. For each param, state: type, valid values/range, format, and when it's needed. Example: 'stages' → 'Comma-separated analysis stages to run: ocr, colors, quality, faces, edges. Default runs all. Omit to skip expensive stages.'
• Split arachne_extract into separate tools: extract_youtube, extract_audio, extract_video, extract_pdf, extract_docx, extract_image. Each tool does one thing; agents compose them as needed.
• Document output schemas for all tools. Include field names, types, and which fields are required for chaining to downstream tools. Example: arachne_search should return [{ url, title, snippet, domain }] with total_count and next_cursor for pagination.
• Remove or heavily gate arachne_torrent_extract. The description hints at copyright violation. If retained, add explicit legal disclaimers and permission checks.