sh-guard-mcp exposes two well-named, security-focused tools with clear intent. Both tools have detailed descriptions (150+ chars) and comprehensive input schemas with proper types and enums. However, output schemas are not documented in the source code, the tool descriptions mention return fields (risk_score, risk_level, intent, MITRE ATT&CK mapping) but no formal schema is visible. Parameter descriptions are present and specific (e.g., 'shell type (bash or zsh), affects zsh-specific rule detection'). Error handling and recovery guidance are not evident in the provided code. The tools follow single-responsibility principle well (classify vs batch), but lack output schema documentation and error categorization.

• Document output schemas for both tools. Define the structure of risk_score (number 0-100), risk_level (enum: safe|caution|danger|critical), intent (string), target (string), data_flow (object), and mitre_techniques (array of strings). Include example responses.
• Add error handling section to tool descriptions. Specify: 'Returns error if command is empty, unparseable, or exceeds 10KB. Error includes the invalid input and suggests valid alternatives.'
• Add 'when to use' guidance. E.g., 'Use sh_guard_classify for single commands before execution. Use sh_guard_batch to analyze multiple commands in one call (faster than sequential calls).'
• Specify limits for sh_guard_batch: 'Accepts 1 - 100 commands per request. Returns results in the same order. If a command fails to parse, that result includes an error field instead of risk data.'
• Migrate from STDIO to HTTP+SSE or Streamable HTTP transport to enable remote access and integration with hosted MCP clients.