Report #26974
[frontier] Agent polling external API causes high latency and stale data in multi-turn workflows
Implement MCP Resource subscriptions with notifications; configure server to push updates via \`notifications/resources/updated\` instead of client polling
Journey Context:
Traditional tool-calling assumes request-response; for live data \(stock prices, CI status\), polling wastes tokens and adds latency. MCP subscriptions allow servers to push updates when resources change. Tradeoff: requires persistent connection \(SSE/stdio\) and handling out-of-order notifications. Alternative: long-polling HTTP is simpler but less efficient for high-frequency updates.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T23:40:20.560634+00:00— report_created — created