Single-tool MCP server providing shell command security assessment. The tool has a clear, specific name (assess_command) and a detailed description. Input schema is present with proper typing. However, the output schema is NOT documented in the visible source code, the agent.rs file shows only input parameter definition and mentions 'RiskAssessment' as a struct but does not expose its field definitions, types, or structure to the MCP client. This is a critical gap because downstream tool chains cannot predict the response shape. The tool follows good naming conventions (verb_noun) and has a security-focused use case with appropriate error handling logic mentioned (severity thresholds, deny lists), but lack of documented output schema and missing parameter descriptions for response fields prevent a higher score.

• Document the RiskAssessment output schema with all fields, types, and descriptions. Example: {matched_rules: string[], severity: 'critical'|'high'|'medium'|'low', alternatives: string[], allowed_for_agents: boolean, explanation: string}
• Expand the 'command' parameter description to: 'A shell command string to assess for security risks. Supports bash/sh syntax including pipes, redirects, and variables. Max 10,000 characters. Must be a syntactically valid shell command or will return a parse error.'
• Add an error handling section to the tool description explaining: 'Returns error if command is syntactically invalid. Returns severity=critical with allowed_for_agents=false if matched against deny rules. Agent should not execute denied commands.'
• Document the decision logic for 'allowed_for_agents': 'True if all matched rules are below severity threshold (configurable per agent). If false, the command should not be executed by the agent without human review.'
• Add examples of response structure in documentation: '{matched_rules: ["rm_without_confirmation"], severity: "high", alternatives: ["use 'rm -i' for interactive mode"], allowed_for_agents: false}'
• Consider adding parameters for threshold tuning: 'severity_threshold: "low"|"medium"|"high"|"critical"' to let agents configure permissiveness per context