This is a red-team lab demonstrating MCP attack vectors, not a production tool server. Both tools have minimal descriptions (under 100 chars), lack parameter descriptions entirely, and schemas are present but incomplete. The 'style' parameter in format_text has no type constraint or description. The 'words' parameter in summarize lacks description. No error handling guidance, no output schema documentation, and no security considerations beyond the intentional attack payloads. Tool names are generic (format_text, summarize) without clear action verbs. This server is deliberately hostile, it injects malicious instructions into the discovery surface via multiple attack modes (override, volume, cache, drift). While this serves a legitimate red-team purpose, it scores extremely low on production quality metrics.

• Add parameter descriptions to both tools. For format_text, document 'style' with valid options (e.g., 'Style to apply: one of bold, italic, uppercase, lowercase, title_case'). For summarize, document 'words' with range (e.g., 'Target summary length in words (10-500)').
• Add type constraints and enums. Define 'style' as an enum with explicit valid values. Define 'words' as integer with minimum=10, maximum=500.
• Document output schemas. Specify what format_text returns (e.g., {type: 'object', properties: {formatted_text: {type: 'string'}}}). Specify what summarize returns (e.g., {type: 'object', properties: {summary: {type: 'string'}, word_count: {type: 'integer'}}}).
• Add error handling guidance. Include descriptions like 'If style is invalid, returns error with list of valid options. If words is out of range, returns error with min/max bounds.'
• Extend tool descriptions to 50-200 characters. Current descriptions are too brief. Add context: 'Format a block of text in the requested style (bold, italic, uppercase, lowercase, title_case). Useful for preparing text for display or emphasis.'