How do you make agents come back to finish long-running MCP jobs after they complete?
agentrsdg · reddit · 2026-09-09
The poster keeps hitting the same wall: tools that kick off long-running jobs return a task id instead of results. The jobs finish fine, but the agent often moves on as if nothing is pending, or checks back having lost so much context that it reruns the next step.
The real question: is something external waking the agent back up, or does the agent handle that itself? The author reports the self-managed approach hasn't worked at all. A practical agent-orchestration pain-point discussion.
More from coding & agent
- Letting Opus 5 touch your codebase without strict guidelines pollutes every assistant's style — intellectronica · 2026-09-09
- fxhoudinimcp: An Open-Source MCP Server Brings 188 Tools for AI Control of Houdini — petewoodbridge · 2026-09-09
- Dev juggling five AI coding harnesses asks: how does anyone manage this? — 0xsachi · 2026-09-09
- One App's User Reviews Expose Vibe Coding's Gap: Shipping Needs Real Engineering — bendee983 · 2026-09-09
- Ink & Switch unveils composable malleable computing; Reinhardt says LLMs still can't refactor it — Ben_Reinhardt · 2026-09-09
- Bot Mesh launches: a social network where agents verify with keys, sign posts, and pay each other in USDC — Daniel_Farinax · 2026-09-09