AgentQueueMcp has well-structured tool definitions with clear naming (verb_noun pattern: send_task, send_result, get_messages, ack_message, list_agents) and comprehensive parameter descriptions. All 6 tools have input schemas with typed parameters and descriptions. However, output schemas are not formally documented, tools return JSON strings rather than structured objects with declared fields. Error handling is minimal; tools lack recovery guidance and actionable error messages. The server uses Azure Queue Storage as a message broker, which is a valid pattern, but the tool descriptions could better explain the at-least-once delivery semantics and idempotency requirements.

• Document output schemas for all tools. Define the structure of returned JSON objects (e.g., send_task returns {sent: bool, taskId: string, conversationId: string, to: string, archived?: string}). Use JSON Schema or a structured format in tool descriptions.
• Add error classification and recovery guidance. For example: 'Message too large' → 'Retryable: false. User-fixable: true. Shorten the payload and retry.' Include specific constraints (e.g., max message size after base64 encoding is 60KB).
• Clarify idempotency semantics in get_messages and ack_message descriptions. Explain that taskIds must be tracked to avoid duplicate processing, and that ack_message is idempotent (calling it twice with the same messageId is safe).
• Add optional confirmation step for ack_message. Consider a dry_run parameter or a separate confirm_ack_message tool to prevent accidental message loss.
• Enhance list_agents description to explain the discovery mechanism (queue names matching the prefix pattern) and when to call it (e.g., before send_task if the target agent name is unknown).

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

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