Critical failures across all dimensions. Tool definitions are inferred from a React UI component (ToolCard.tsx), not from actual MCP server code. No explicit tool registration visible in source. Descriptions are minimal (10-40 chars, well below 50-200 char baseline). Input schemas lack type definitions and constraints, parameters are listed but not formally typed. No output schemas documented. No error handling guidance. The 'bash' and 'agent' tools are irreversible but lack confirmation patterns. Multiple tools accept alternative parameter names (file_path vs path, command vs cmd) without documenting mutual exclusivity. No pagination, no idempotency markers, no permission gates.

• Locate and audit the actual MCP server implementation (likely in mcp-server/ directory). Extract tool definitions from the server code, not from the UI component. Verify each tool is registered with explicit name, description, input schema, and output schema.
• Expand all tool descriptions to 50-200 characters. Include: what the tool does, when to use it (vs similar tools), what it returns, and any prerequisites. Example: 'Read file contents from disk. Use for source code inspection, config review, or log analysis. Returns file text or error if not found or unreadable.'
• Add formal JSON Schema type definitions to all parameters. Use 'type': 'string', 'type': 'number', etc. For constrained values, add 'enum' arrays. For alternative parameter names, document mutual exclusivity in descriptions: 'Either file_path or path (not both).'
• Document output schemas for all tools. Specify returned fields, types, and structure. Example for read: {type: object, properties: {content: {type: string}, size: {type: number}, encoding: {type: string}}}.
• Add confirmation/dry-run patterns to bash and agent tools. Implement a 'confirm_bash_execution' or 'preview_agent_delegation' step before irreversible actions. Return actionable error messages: 'Command would delete 500 files. Confirm with confirm_bash_execution(command_id=..., confirmed=true).'