GmailMCPServer has only 1 tool (send_email) with a basic but incomplete schema. The tool has a verb-based name (send_email) and minimal parameter descriptions. The schema is visible and typed, but descriptions are terse (under 20 chars for most fields) and lack actionable guidance. The tool does not explain when to use it, what happens on error, or recovery paths. No idempotency guidance, no rate-limit documentation, no dry-run/confirmation pattern despite being a write operation. The resource (read_recent_inbox) is present but tool-specific scoring focuses on send_email. Overall, this server provides basic structure but falls short of production-grade quality for an agent-facing tool.

• Expand send_email description to 150-200 characters: 'Sends an email via Gmail to the specified recipient. Returns success confirmation. Requires valid email address format. On failure (invalid email, auth error, rate limit), returns error details, use send_email again with corrected parameters or wait before retrying.'
• Add parameter descriptions: 'to' → 'Recipient email address (required, must be valid email format, e.g. user@example.com). If invalid, the tool returns an error.'; 'subject' → 'Email subject line (required, 1-998 characters). Supports Unicode.'; 'body' → 'Email body text (required, plain text or simple HTML, up to 25 MB). Line breaks are preserved.'
• Add input validation: validate 'to' against email regex; reject empty 'subject' or 'body'; return structured error: {'error': 'Invalid email address', 'value': <to>, 'constraint': 'Must match pattern user@domain.com'}
• Implement retry guidance: catch Gmail API exceptions (rateLimitExceeded, authenticationFailure, etc.) and return 'Temporary error: rate limit reached. Wait 60 seconds and try again.' or 'Auth failed: token may be expired. Regenerate token.json with get_token.py.'
• Add idempotency check or dry-run mode: return {'status': 'success', 'recipient': <to>, 'subject': <subject>, 'message_id': <gmail_message_id>} instead of plain string, enabling agents to verify the message was actually sent.
• Credential handling: check if token.json exists at startup; if missing, raise clear error: 'token.json not found. Run: python get_token.py' instead of failing silently during the first send_email call.