Four tools with mixed quality. mcp_tool_search, mcp_resource, and mcp_prompt have complete input schemas (Zod-validated) and descriptions, but descriptions are brief (10-60 chars) and lack actionable context for LLM selection. clarify has a complex nested schema with good structure but minimal description (58 chars). No output schemas documented. Error handling present in mcp_resource but absent in others. Tool names follow verb_noun convention. Parameters mostly typed but lack detailed constraints and dependency documentation.

• Expand tool descriptions to 100-200 chars. For mcp_tool_search: 'Search MCP tool catalogs by keyword. Returns matching tools with schemas. Use this to discover available capabilities before invoking them. Results are materialized for immediate use.'
• Document output schemas explicitly. Define return types for each tool: mcp_tool_search returns {matches: [{callableToolName, description, inputSchema, materialized, name, serverId}], servers: [{serverId, toolCount, resourceCount, promptCount, discoveryError, instructions}]}.
• Add parameter constraints and examples. For 'query' in mcp_tool_search: 'Search string (e.g. "email", "calendar"). Partial matches supported. Empty string returns all tools.' For 'arguments' in mcp_prompt: 'Object mapping template variable names to values (e.g. {"user_id": "U123", "format": "json"}).'
• Implement error handling for mcp_tool_search and mcp_prompt. Return structured errors with recovery guidance: 'No tools found. Try a broader query or list all tools with query="".' 'Prompt not found. Call mcp_tool_search to discover available prompts.'
• Document the dynamic tool registration flow. Clarify whether materialized tools are cached, how agents invoke them, and whether repeated searches are idempotent. Add a note: 'Calling mcp_tool_search with the same query returns the same results and does not re-register tools.'