Four tools with strong naming (verb_noun pattern), comprehensive descriptions (194 - 500+ chars), and detailed input schemas. upload_video and edit_video_metadata have excellent parameter documentation with constraints, enums, and dependency hints. list_channels and verify_channels are simpler but well-described. Output schemas are documented (UploadOutput struct visible). However, output schema documentation is embedded in Go structs rather than exposed in JSON Schema; parameter descriptions rely on jsonschema tags rather than formal JSON Schema enum/pattern constraints; error handling guidance is present but not formalized in tool definitions.

• Expose output schemas as JSON Schema in tool definitions. For upload_video, declare UploadOutput fields (status, video_id, stage, category, error, etc.) with types and descriptions in the tool's outputSchema property.
• Formalize parameter constraints: declare privacy_status as enum [public, unlisted, private]; category_id as pattern ^\d+$; tags as maxItems 500; title as maxLength 100. Use JSON Schema properties, not just descriptions.
• Add tool annotations: mark upload_video with destructiveHint=true and idempotentHint=true; mark list_channels and verify_channels with readOnlyHint=true. This guides LLM planning and safety checks.
• Document error responses as part of the tool contract. Define a structured error schema (stage, category, error, hint, rolled_back) and include it in tool documentation so LLMs know what to expect on failure.
• Remove example values from descriptions (e.g., 'e.g. 28 = Science & Technology'). Instead, reference the YouTube category taxonomy or provide a discovery tool (list_categories) that returns valid IDs.