Single tool 'fetchKGTools' has a well-structured input schema with 7 parameters, all typed and described. However, the tool definition is inferred from a React component file (ToolsBrowserDrawer.tsx), not from explicit MCP server registration code. No output schema is documented. The description is adequate (89 chars) but lacks guidance on when to use this tool vs alternatives, what structure is returned, or error recovery. The tool name 'fetchKGTools' uses a verb but is somewhat generic ('fetch' could mean HTTP GET, database query, or cache lookup). No evidence of error handling, validation, or response shaping in the provided source.

• Add explicit MCP tool registration in the Python server code with complete input and output schemas. Move tool definition from React component to server-side registration.
• Document the output schema: return type (array of objects?), fields per tool (id, name, description, risk_level, domain, function, exposure?), and total count or pagination cursor.
• Expand description to 100-150 chars: 'Fetch tools from the knowledge graph by operation. Returns matching tools filtered by exposure level, domain, function, and risk. Use this to discover available tools before planning a pipeline.'
• Add error handling guidance: 'If operation is not found, returns empty list. If domain or function filter is invalid, returns 400 with valid enum values.'
• Rename tool to 'search_kg_tools' or 'list_kg_tools' to clarify intent and align with verb_noun naming convention.