Four tools with significant definition gaps. mcp_search_tools and mcp_get_tool_details have reasonable descriptions (60-80 chars) and basic schemas, but mcp_execute_code and mcp_code lack meaningful differentiation and have minimal descriptions (under 100 chars). All tools accept only string/number inputs with no enums, ranges, or validation constraints. No output schemas documented. Error handling guidance absent. Tool naming uses underscores but lacks clarity on why two nearly identical code execution tools exist (mcp_execute_code vs mcp_code). No parameter descriptions beyond bare type info.

• Merge mcp_execute_code and mcp_code into a single tool with clear naming (e.g., 'mcp_execute_javascript'). Document the namespace, available APIs, and return value format in the description.
• Add output schemas for all tools. mcp_search_tools should return {tools: [{name, description, risk}], total_count, has_more}. mcp_get_tool_details should return {name, description, schema, typescript_declaration}.
• Add parameter descriptions: 'limit' (1 - 100, default 20), 'query' (free-text search, case-insensitive), 'code' (JavaScript source; must not use require/import; returns last expression value or console output).
• Document error handling: 'Code execution errors return {error, message, stack}. Syntax errors are retryable; runtime errors may indicate invalid tool calls, use mcp_get_tool_details to verify schema.'
• Add safeName parameter description: 'The safe name of the tool (from mcp_search_tools results). Use this exact value to avoid lookup failures.'