Server has 3 tools with significant definition gaps. Two Python tools (get_translation_prompt, get_answer_prompt) are prompts, not tools, they return static text templates, not actionable results. The curl tool has a basic schema but lacks error handling guidance. Descriptions are present but generic. No parameter validation, no output schemas documented, no error recovery patterns. Tool composition is weak: prompts should be separate from tools. STDIO transport limits remote accessibility.

• Move get_translation_prompt and get_answer_prompt to the prompts capability (already correctly decorated with @mcp.prompt()). Do not expose them as tools. Tools should perform actions, not return static templates.
• Add structured output schema for use_curl: return {status_code: int, headers: {}, body: string, content_type: string, size_bytes: int} instead of raw HTML. Document this in tool description.
• Enhance use_curl error messages with recovery guidance: 'Connection timeout (30s). Retry with a shorter timeout or check if the URL is accessible.' Include error classification (retryable vs fatal).
• Expand use_curl description to 100-200 chars: 'Fetch and return the raw HTML content of a webpage. Returns status code, headers, and body. Use when user explicitly requests to view a webpage. Supports HTTP/HTTPS only. Max 10MB response.'
• Add input validation to use_curl: validate URL scheme (http/https only), enforce timeout (30s max), add SSRF protection (block private IP ranges), and return clear validation errors.