FWIW VSCode still has this feature. I’m in the process of evaluating cursor and this may be a deal breaker. Commands seems like a good tool but custom modes are way more powerful
Also you can add rules and scope them to specific directories, file types, etc.
Not the same for sure but still helpful
Come on everyone get voting on the feature request to return a feature that already existed, was popular with power users, got removed without our request, and now needs to be requested back in.
Return the Custom Modes features
We cant let the bad good guys win!
what kinda dumbass practice is this? introduce a feature, wait a month, ■■■■■■■ nuke it??? yeah this is not an enterprise company or product
I upvoted the feature request. Used this feature and shared it’s usefulness at my company.
I had a “Testing” agent that had a specific prompt about how to write tests according to our stack and with certain mcps enabled for writing tests. Enjoyed swapping to it and seeing the LLM do a much better job than when I used the typical Agent mode.
Soon on Netflix 
This smells very fishy to me personally. They shift over to usage only billing, and then lock us out of the capabilities that allow us to more effectively manage our tools and therefore context usage. If they want wide spread team adoption they actually need to support the ability of teams to mitigate and manage misuse. Commands dont do that, and I dont think they ever will. The ability to not only set the system message of the agent, but also tool use and default agent model are all foundational functionality that gives me as a manager piece of mind as to the reliability of the results from those custom agents. I was using my own Plan agent months before we ever saw it in cursor, as well as Research, Reflect, Debug, and Unit Testing agent modes, each geared to work in tandem with one another within a standardized workflow. I see cursor trying to roll out their own equivalents within the system to rack up some more usage. Fine, fair, more power to them.. but guess what guys… that will only work as long as the code stays maintainable, as all the very smart people on this thread seem to know. The more control they take away from the ai enabled IDE, honestly, the less reason to use Cursor… because that is what in the end makes it worthwhile. The devs missed the mark badly here.
Cancelled my subscription, Vote with your wallets.
One note here, getting my users locked into specific models for specific modes really helped keep them off the pricier ones by accident… which really is a factor for newer users.. Hate all the team usage being sucked up by accident…
Agreed. Enterprise implementations want a level of consistency for their teams. modes gave my team manager piece of mind at a minimum and a vital repeatable structure as we started developing it into our own AI-DLC. This kind of a change makes me question if i can trust this IDE to be reliable in the future. I have been excited about all that can be done with a true AI enabled IDE. But if functionality is just going to be dropped arbitrarily… I cant look to my leadership with any confidence about estimates, budgets , etc because the workflows and processes that I established in order to support a larger deployment are now defunct… That ■■■■■.
We need ability to disable tools
I’m in agreeance @condor. I used the custom modes a lot as I can get the agent to perform longer more autonomous tasks easier. I’m also on the ultra plan and spend more on top of that. Its very disappointing a feature I was activity using was pulled.
Bring back custom modes and expand their functionality, do not remove them.
There are basically two things that permit Cursor to get away with almost everything:
Big Provider’s expensive plans (like ChatGPT Pro, Claude Max) also subsidize power users, but don’t have the flexibility of model switching. And all other “neutral” coding vendors (from Jetbrains to Kiro to Windsurf) don’t offer that sweet 50% subsidy.
Perhaps that big piles of VC money that Cursor is burning through and we benefit from causes a sort of “resource curse” for Cursor, when they can get away with such bad product quality management and “move fast and break things”-style development at its worst that users tolerate only because of that unreasonable price advantage that they get.
I’d like to share my use case for the Custom Modes feature, as I’m currently using version 1.7 (because the lack of that feature in 2.x). I’ve created two custom modes: Code Review and Merge (Pull) Requests. I use these modes to streamline my workflow, especially since I often work with multiple Git repositories per project.
The custom modes help me automatically scan my project directory (to check if I use single git repository or multiple), gather the diff changes between committed and uncommitted changes, and compare my working branch with the “parent” branch. It also helps generate commit messages that follow Git flow rules, including a ticked number based on branch name for easy reference. Additionally, the modes sum the time I’ve worked on the branch based on commit messages, and at the end, I can squash the commits and generate a clean, professional summary for business analytics.
The Custom Modes feature has been a powerful tool in this process. It’s as simple as selecting the mode from a dropdown, adding a few extra prompts, and everything works seamlessly. It really streamlines the entire process, saving me significant time and effort.
Custom Modes also allowed me to save commands as internal prompts for the agents, and gave me the flexibility to add a traditional prompt to describe a specific use case. While the UI could definitely use some improvement, I was hoping that this feature would be enhanced rather than removed entirely. One great addition would be the ability to select which model should be used to execute a particular mode, since some models are better suited for tasks like creating commit messages and summaries than others.
Hey, for a while now there was no feedback on why there is no Vertex AI integration… The changelog was lacking big time for months. There should be main and alpha channel on changelog page.
Now removing custom modes without any user discussion… This would be fine for open source project with daily releases (like OpenCode), not for a product with X-XX k $ /month price.
I am iOS Software Engineer and we need to use extensive MCPs to allow LLMs to interact with xcode build system. This causes us to either bloat the context window with ~100k token of tools, or, build specialised agents (modes). You want to tell me, now I have to edit what MCPs and tools are enabled before creating a new chat, not being able to work in couple of modes simultaneously?
Why custom modes are more valuable over commands:
You say the tool specification was not common in custom modes. Is that an argument to remove the feature? I don’t think you’ve come up with proper conclusion based on your data.
Can anyone from Cursor team tell us, why did you remove the custom modes, instead of iterating on the feature? Especially seeing how multi-agentic approach becomes more and more common? I would like custom modes not only to come back, but iterated on. I want the modes to have an option for separation of contexts. I want orchestrator mode that can execute the plan on a high level and based on mode description delegate planned tasks to proper modes (sub-agents), awaiting response (finished task). I want to effectively reduce context bloat / pollution and have more granular control over the flow. MCP was a big hit, but it grew too fast. The future is in A2A protocol and RAG-like approach with MCP tools being used to very limited range of use cases. Look at OpenCode, Claude Code, Roo Code and more. Agent with sub-agents, orchestrator, skills - this is the selling point right now.
This. Exactly this! Not to put too fine a point on it I see Cursor as having 2 customer bases fundamental to their success.
The vibe coders that will probably never write a full app on their own
the old hats that have written code across stacks and can see what is necessary to develop at scale, with repeatable methodologies, while operating within business based constraints and prerogatives.
The first group, are going to figure out they need things repeatable, and maybe cursor will eventually be able to wrap up and deliver that whole experience so they don’t even need a editor view…
But, that isn’t today, and cutting off a feature (allbeit a beta one) that had wide-ranging flexibility to your target architect/engineer/developer audience without a meaningful alternative doesnt give the best feels all around.
This cant be serious! I would never expect this could happen. All my workflow is broken. If you are not putting back the custom modes I am cancelling my subscription.
Plenty of options out there now. Amazon and Google have entered the IDE race.
I switched back to cursor 2.0 to keep using custom modes but I see that sonnet 4.5 is not accessible from cursor 2.0 anymore while totally accessible from 2.1
Can someone from cursor tell what is wrong with having a setup with predefined modes with different mcps assigned to them? I am really curious.
Apparently, per a previous response, the devs assessed that tool lock down on custom agent modes “was not statistically significant”, so they assumed the feature wasnt a winner.. Versus perhaps going a step closer and recognizing that the users that were/are using that feature are generally power users building on top of their product. Or at least, that is my hope, versus it being something more nefarious like they don’t want power users in support of a agentic first approach (not that the two have to be mutually exclusive) .