Server defines 4 tools with complete JSON schemas and descriptions. Tool names follow verb_noun pattern (delegate_to, web_task_*). Descriptions are adequate (100-200 chars) and explain purpose clearly. However, parameter descriptions lack detail on constraints, formats, and error recovery. Output schemas are not documented. Error handling is minimal, no guidance on retryability or recovery steps. Missing tool annotations (readOnlyHint/destructiveHint). The delegate_to_ds_web tool has a complex async contract (blocks until completion) that could benefit from clearer timeout and failure mode documentation.

• Document output schemas for all tools. Example: web_task_claim returns {taskId: string, instructions: string}; web_task_list returns {tasks: [{id, status, label}], total: number}.
• Add constraint descriptions to parameters: 'prompt' (max 10000 chars, self-contained instruction), 'label' (optional, max 100 chars), 'taskId' (UUID format).
• Enhance error recovery guidance: 'If delegate_to_ds_web times out, call web_task_list to check if the task is still queued, or restart the delegate and retry.'
• Add tool annotations: mark web_task_claim and web_task_list with readOnlyHint=true; mark web_task_settle with destructiveHint=true (modifies task state).
• Clarify timeout behavior: document default taskTimeoutMs, whether it can be overridden per-call, and what the LLM should do on timeout (e.g., 'Timeout after 15 minutes. Check delegate status or retry with a simpler prompt.').

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

• Relies on Stateful initialize / Mcp-Session-Id (removed; protocol is stateless) - make each request self-contained