This is a deliberately malicious demo server (evil-mcp) designed to showcase attack vectors that wardn blocks. All four tools expose dangerous capabilities (read_secret, exfiltrate, nuke, shell_exec) with minimal safety guardrails. While schemas are present and descriptions exist, the server is intentionally unsafe and unsuitable for production use. Tool descriptions are extremely brief (10-50 chars), parameter descriptions are missing entirely, and there is no error handling, validation, or recovery guidance. The server is STDIO-only, making it inaccessible to remote clients.

• Add detailed parameter descriptions to every input field. Example: 'path: Absolute or home-relative path to read (e.g. ~/.ssh/id_rsa or /etc/passwd). Must be readable by the current user.'
• Expand tool descriptions to 100-200 characters explaining WHAT the tool does, WHEN to use it, and what it returns. Example for read_secret: 'Reads and returns the contents of a file from the user's home directory or absolute path. Use to retrieve configuration, keys, or data files. Returns file contents as plain text (first 200 chars).'
• Add enum constraints or regex patterns to parameters. Example: 'url must be a valid HTTPS URL (https://...)' for exfiltrate.
• Implement error handling with actionable messages. Return structured errors with code, message, and recovery hint. Example: '{error: "path_not_found", message: "File not found at /home/user/.ssh/id_rsa", hint: "Check the path exists and is readable."}'
• Add confirmation/dry-run support for destructive tools (nuke, shell_exec). Return a preview of what would happen and require explicit confirmation before executing.