Skip to content

sub-agents in background with MCP tools are flaky #2357

@aheritier

Description

@aheritier

I have a root agent with sub-agents configured to be executed in background and these sub-agents have various MCP servers.

Often they never succeed to come back with an answer.

✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_iv
Status:  running
Runtime: 4m29s

--- Output ---


[No new output since last check — poll #11]
✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_ix
Status:  running
Runtime: 4m29s

--- Output ---


[No new output since last check — poll #11]
✓ Search Files Content  . ([REDACTED]) no matches
✓ Search Files Content  . ([REDACTED]) no matches
✓ Stop Background Agent  task_id=agent_task_[REDACTED] Background agent task agent_task_[REDACTED] stopped.
✓ Stop Background Agent  task_id=agent_task_[REDACTED] Background agent task agent_task_[REDACTED] stopped.

root


The previous Notion and Granola agents stalled. Let me retry both with fresh dispatches — this time I’ll try a direct page ID lookup for the Notion inventory.

✓ Run Background Agent
agent:
mark_iv

expected_output:
The full repository inventory table with repo names, status/decisions (merge, keep, archive), and any notes.

task:
Find and retrieve the Notion page titled "[REDACTED]" (page ID: [REDACTED]). Return the FULL content — especially any table or database listing repositories, their status (merge/keep/archive), ownership, and notes. URL: [REDACTED]
Background agent task started with ID: agent_task_[REDACTED]
Agent: mark_iv
Task: Find and retrieve the Notion page titled "[REDACTED]" (page ID: [REDACTED]). Return the FULL content — especially any table or database listing repositories, their status (merge/keep/archive), ownership, and notes. URL: [REDACTED]
✓ Run Background Agent
agent:
mark_ix

expected_output:
Full transcript or notes content from the retro, specifically any mention of repositories to merge into docker/ai.

task:
Retrieve meeting notes/transcript from Granola: [REDACTED] . I need the full content, especially any discussion about which backend repositories should be merged. Return everything you can extract.
Background agent task started with ID: agent_task_[REDACTED]
Agent: mark_ix
Task: Retrieve meeting notes/transcript from Granola: [REDACTED] . I need the full content, especially any discussion about which backend repositories should be merged. Return everything you can extract.
✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_iv
Status:  running
Runtime: 3s

--- Output ---

✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_ix
Status:  running
Runtime: 3s

--- Output ---

✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_iv
Status:  running
Runtime: 7s

--- Output ---


[No new output since last check — poll #2]
✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_ix
Status:  running
Runtime: 7s

--- Output ---


[No new output since last check — poll #2]
✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_iv
Status:  running
Runtime: 10s

--- Output ---


[No new output since last check — poll #3]
✓ View Background Agent  task_id=agent_task_[REDACTED]
Task ID: agent_task_[REDACTED]
Agent:   mark_ix
Status:  running
Runtime: 10s

--- Output ---


[No new output since last check — poll #3]

Metadata

Metadata

Assignees

Labels

kind/bugSomething isn't working

Type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions