Single tool 'fetch' has a well-structured schema with all parameters typed and described, and includes a meaningful tool description. However, it lacks critical production-grade error handling guidance, output schema documentation, and security hardening. The tool accepts user-provided URLs without documented sanitization against SSRF or injection attacks. Error messages are generic ('Error fetching URL') rather than actionable recovery guidance. No pagination or result limiting is documented despite potentially large content responses. The server demonstrates basic correctness but lacks patterns expected of production tools.

• Add explicit output schema documentation to fetch tool description: 'Returns: string, markdown-formatted or raw HTML content, truncated to maxLength characters starting from startIndex.'
• Enhance error categorization. Instead of generic try-catch, return structured errors: (1) 'robots_blocked' with message 'Site forbids fetching via robots.txt. Try a different source.', non-retryable. (2) 'timeout' with 'Connection timed out after 30s. Retry in 1 minute.', retryable. (3) 'http_error_404' with 'URL not found. Verify the URL is correct.', user-fixable.
• Add SSRF protection: document that internal IPs (127.0.0.1, 10.0.0.0/8, 169.254.x.x, 172.16.0.0/12, 192.168.0.0/16) and known metadata endpoints (169.254.169.254, localhost) are blocked. Code review should verify these are actually blocked server-side.
• Document rate limiting policy: 'Tool is rate-limited to 10 requests per minute per source. Subsequent requests within 60s will be queued or return 'rate_limit_exceeded' error.'
• Expand parameter descriptions: (1) maxLength: 'Maximum characters to return (1-1000000). Defaults to 5000. Large values may cause timeout or context window exhaustion.' (2) startIndex: 'Character offset to begin extraction (0-based). Useful for paginating large documents. If startIndex >= content length, returns empty string.' (3) raw: 'If true, return raw HTML. If false (default), convert to markdown for readability.'
• Add security note to description: 'For security, this tool blocks internal IP ranges and cloud metadata endpoints. It respects robots.txt. Credentials should never be in URLs, use environment-based auth instead.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry