Where does the bug appear (feature/product)?
Cursor CLI
Describe the Bug
I’m working on creating a Sandbox environment that includes multiple repositories and instruments available on the machine, so that the worker could navigate through the code-base of various repositories, not just the primary-repo.
Once the environment is ready - agent worker is being spawned with
agent worker start --worker-dir /workspace/self-hosted-agent
--management-addr :8080
--single-use
--name agent-worker-123
--data-dir /workspace/.cursor/agent-data
--labels-file /workspace/.cursor/worker-labels.json
--api-key crsr_123
Once agent worker has started - it would normally appear in Dashboard/Cloud Agents with status Idle which would allow selecting self-hosted workers at /agents page and start the task.
Up until yesterday 3PM CET - this same exact flow would normally work with no issues, meaning that once Self-Hosted worker is being selected and a task being prompted - a worker would start executing it returning with feedback. Followups would work too.
Since that time - I’ve noticed some changes in Cursor Web App, mainly the Self-Hosted agents block was not displayed anymore in Dashboard/Cloud Agents and instead of selecting Self-Hosted Agents we could now select a certain worker in the drop-down which is convenient.
However - now when the task is being prompted to the worker it first goes through Setup the environment step than jumps to Waiting for worker and fails after 3 attempts (Screenshot).
By looking at logs of the worker - after the initial block of logs which seems usual - I’m only seeing heartbeat debug logs every 30 seconds and nothing else. The initial logs look good, a management server is being spawned, it can be queried and the heartbeat logs make me think that the HTTPS connection is fine.
I’ve made sure I don’t have any workers sitting in Idle or Running state, tried to delete all agents that I had by this time using the Agents API but so far no luck.
I’ve also tried to logout everywhere, revoke tokens and re-issue them again.
I’ve tried to revoke and re-apply GitLab MCP connection as well.
I’ve read Self-hosted worker stuck on "Waiting for self-hosted worker" after enabling/disabling MCP toggle in cursor.com/agents but so far no luck getting unstuck here.
Steps to Reproduce
- Spawn a worker
- Prompt a Task
- Fails with Unknown Error
Expected Behavior
- Spawn a worker
- Prompt a Task
- Executes the task
Screenshots / Screen Recordings
Operating System
Linux
Version Information
CLI Version: 2026.03.30-a5d3e17
Model: Auto
OS: linux (x64)
For AI issues: which model did you use?
Composer 2
For AI issues: add Request ID with privacy disabled
Worker ID: bc-440a368f-97bc-494f-88cd-e8e918f064f0
Additional Information
LOGS:
10:27:02.272 DEBUG Setting host header for ALB routing ctx=worker-mode meta={hostHeader: “api2.cursor.sh”}
10:27:02.274 INFO Session state after connect ctx=worker-mode meta={state: “connecting”}
10:27:02.274 INFO Starting to iterate over stream frames ctx=worker-mode
10:27:02.274 INFO Starting management server ctx=worker-mode meta={address: “:8080”}
10:27:32.690 DEBUG Received frame ctx=worker-mode meta={frameCount: 1}
10:28:02.686 DEBUG Received frame ctx=worker-mode meta={frameCount: 2}
10:28:32.687 DEBUG Received frame ctx=worker-mode meta={frameCount: 3}
10:29:02.687 DEBUG Received frame ctx=worker-mode meta={frameCount: 4}
Does this stop you from using Cursor
Yes - Cursor is unusable