Hi everyone,
I’m a paid Cursor user who primarily uses Cursor Mobile for Cloud Agent work. Overall the mobile experience is great and being able to run agents on the go is genuinely useful. That said, I’ve hit several bugs and have some suggestions that I think would help a lot of mobile users. Posting here in case others have seen the same things.
My setup: Cursor Mobile on two devices. iPhone 16 Pro running iOS 27 beta 5. iPad Air 7 M3 running iPadOS 26.6. Personal Gmail account. Typical workflow is no repo selected so cloud VM mode. Example run: https://cursor.com/agents/bc-01a042a3-0af7-724f-8f17-1be85dc70c6d. Most of these issues occur on both devices unless otherwise noted.
Bug 1, tasks end abnormally, possibly context limit without compression: When running Cloud Agent without a local repo in cloud VM only mode, tasks sometimes end abruptly with no clear reason. My guess is the context window limit is being hit. On mobile it seems like the agent doesn’t compress or summarize context the way desktop might, it just kills the task. Impact is lost progress, worse output quality, and having to start over. Can the team confirm whether mobile handles context limits differently? A warning or graceful compression before termination would be much better than a silent crash.
Bug 2, Continue working hides the thinking trace: After an abnormal termination, tapping Continue working makes the thinking and reasoning process disappear. Normal behavior is a blue chevron you tap to expand the full thinking trace. After continue, tapping worked only shows commands used, or literally no details. Hard to debug or pick up where the agent left off without seeing what it was thinking.
Bug 3, model selection doesn’t persist and Fast silently turns on: Model and effort settings don’t stick between sessions. I keep finding Fast mode auto-enabled even when I didn’t choose it. Worse, when Fast is on there is no Fast label in the chat bar. You only notice if you open the model picker dropdown. Easy to run Fast without knowing it, which may hurt quality. Please persist model and effort selection, and always show the active effort level such as Fast in the chat UI.
Bug 4, final response streams inside thinking block not the chat bubble: Sometimes the main reply text streams inside the thinking section instead of the normal chat area. The body then dumps all at once after thinking finishes, rather than streaming token by token in the message bubble. Desktop seems to separate these cleanly, mobile doesn’t always.
Bug 5, gray divider splits response so you can’t select or copy as one block: In the main reply area, content is sometimes split by a gray horizontal line that looks like a section divider. Text above and below are separate UI elements. You can’t drag one selection across the divider to copy the full message, you have to copy each chunk separately. Painful for long structured replies like lists, code plus explanation, or multi-part answers. Please keep contiguous replies selectable as one block, or add a Copy full message button.
Feature request, thinking UI like desktop: Mobile opens thinking in a separate modal. Would prefer inline expand and collapse with a chevron in the message, same as desktop.
Feature request, setting to expand thinking by default: A toggle in settings for users who always want to see the trace.
Feature request, selectable VM tier with usage-based billing: Let users pick cloud VM specs such as CPU or RAM tier with extra quota cost for heavier workloads.
Feature request, unified usage display on PC and mobile: Desktop shows percentage only, please add absolute token or credit usage. Mobile shows absolute only, please add percentage. Both platforms should show percentage and absolute values consistently.
Feature request, faster or per-conversation usage updates: Mobile usage doesn’t seem real-time and refreshes on some interval. Would love around 60 second refresh, or per-conversation token and credit cost shown after each chat.
Summary: Bug 1 abnormal task end possibly context limit with no compression. Bug 2 Continue working hides thinking trace. Bug 3 model and Fast not persisted, Fast hidden in chat bar. Bug 4 response streams in thinking block not chat. Bug 5 divider prevents selecting and copying full reply. Features include desktop-style thinking UI, default-expand setting, VM tier selection, unified usage display, and faster or per-chat usage updates.
Has anyone else run into these on mobile? Happy to share screenshots, screen recordings, or Request IDs if the team wants to dig in. Thanks!
