This server exposes 10 tools via HTTP but lacks fundamental MCP compliance. Tool definitions are inferred from Express routes rather than explicitly registered via MCP protocol. No input schemas are visible in the code, parameters are extracted from req.body or req.params without formal JSON Schema definitions. Descriptions are present but minimal (10-30 chars), well below the 50-200 char baseline for LLM-optimized tools. No output schemas documented. Error handling returns generic HTTP status codes without recovery guidance. The codebase is a custom Express server, not an MCP server, it does not implement the Model Context Protocol at all.

• Implement MCP protocol: use @modelcontextprotocol/sdk to register tools with proper tool definitions (name, description, inputSchema). Replace Express route inference with explicit MCP tool registration.
• Add formal JSON Schema for all tool inputs. Example for list_containers: {"type": "object", "properties": {"all": {"type": "boolean", "description": "Show all containers including stopped ones (default: true)"}}, "required": []}. Ensure all parameters have type and description.
• Expand descriptions to 80-150 chars with WHEN/WHY context. E.g., 'List all Docker containers managed by OrbStack. Use this to discover available containers before starting, stopping, or inspecting them. Returns container ID, name, status, and image.'
• Document output schemas and strip verbose API responses. For list_containers, return [{id, name, status, image}] instead of full dockerode inspect output. Include pagination (limit, offset, total_count) for large result sets.
• Implement error recovery guidance. Replace 'error: error.message' with structured errors: {"error": "Container not found", "code": "CONTAINER_NOT_FOUND", "recovery": "Use list_containers to find available container IDs."}. Categorize errors as retryable, user-fixable, or fatal.