Terraform Assistant MCP has basic tool definitions but significant quality gaps. All 8 tools have names starting with action verbs (terraform_*) and descriptions present. However, descriptions are minimal (10-30 chars), parameter descriptions are generic or missing, and output schemas are completely undocumented. Error handling returns raw stderr without guidance. No input validation, no enums for constrained values, no pagination support. Tools like terraform_apply and terraform_destroy are high-risk (IRREVERSIBLE/DESTRUCTIVE) but lack confirmation patterns or dry-run support. Parameter schemas exist but lack detail (e.g., 'backend_config' and 'var' are bare objects with no field documentation). This server would require substantial rework before production use.

• Expand all parameter descriptions to 50-150 chars. For 'backend_config' and 'var', document expected field names, types, and examples. E.g., 'backend_config: Object with Terraform backend settings (e.g., {"bucket": "my-bucket", "key": "terraform.tfstate"})'.
• Document output schemas for all tools. Specify whether output is JSON, plain text, or structured. For terraform_plan, return parsed JSON with resource changes, additions, deletions. For terraform_show, return structured state snapshot.
• Add confirmation pattern for terraform_apply and terraform_destroy. Implement a dry_run parameter or a separate confirm_destroy tool. Require explicit approval before irreversible operations.
• Implement input validation: validate working_dir is within allowed paths (prevent traversal), sanitize var values to prevent command injection, validate var_file exists and is readable.
• Enhance error handling: categorize errors (retryable: network timeout; user-fixable: invalid var syntax; fatal: permission denied). Return actionable messages: 'Invalid variable syntax in var_file. Check line X: expected key=value format.'