Three tools with inconsistent quality. All have descriptions (10-50 chars, below 194-char baseline), but schemas are incomplete or missing. mcp.web_search and mcp.supabase_query have partial input schemas (query/limit, operation/table/filters/limit) but no documented output schemas. mcp.heygen_video_agent_generate has minimal schema (only 'prompt' param). No tool declares output structure, pagination, or error recovery guidance. Naming is verb-noun compliant but descriptions are generic and lack WHEN/WHY context. No parameter validation rules, enums, or constraints visible. No evidence of idempotency, rate limiting, or permission gates.

• Add output schema to each tool. E.g., mcp.web_search should document: {"results": [{"title": string, "url": string, "snippet": string}], "total_count": integer}. Include pagination (next_cursor or offset) if results can exceed limit.
• Expand descriptions to 100 - 200 chars. Example: 'Search the public web for information. Returns up to N results with title, URL, and snippet. Use when user asks factual questions or needs current information. Returns empty list if no matches found.'
• For mcp.supabase_query, document filter syntax explicitly. E.g., 'filters: object with column names as keys and values as equality constraints. E.g., {"status": "active", "user_id": 42}. All filters are AND-combined.'
• Add error recovery hints to descriptions. E.g., 'If no results found, try broader search terms or check spelling. If permission denied, verify database credentials.'
• For mcp.heygen_video_agent_generate, add a dry_run parameter (boolean, default false) so agents can preview before committing. Or require explicit confirmation via approval_gate pattern already present in codebase.