Cursor @Docs feature STILL BROKEN

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

This is a long, long standing issue now. Its been two months or more, probably more based on the last time I can remember the docs feature actually working. I relied HEAVILY on the @Docs feature, because without it, the agent is often just guessing or assuming, which is evident in its responses. Even if the agent/model are able to search the web, the results are just not as good as when you can use @Docs.

I have just moved to a brand new Macbook M2 Pro, clean slate, very light weight setup, brand new fresh clean install of Cursor. The @Docs feature still does not work in Cursor. The agents, even though they sometimes seem to be referencing the docs, at least they say they are, they are NOT, and the results of their work without docs is often very pitiful, riddled with guesses, assumption, and blatant fabrication. This was never the case, when @Docs actually worked.

When @Docs worked, you could clearly see the sections of the indexed docs that the model referenced, and the results were phenomenal. The models would know exactly how to use third party frameworks, and could solve complex problems quickly and very effectively. Right now, I’m getting a ton of guesswork and fabrication of “knowledge” that is simply false, and the agent (even after many web searches) cannot build correct code. In this case, I am trying to build a Better Auth server.

I’m sharing two request ids. One with Composer 1, another with Sonnet 4.5. Neither model, seems capable of actually referencing the docs I have indexed for Better Auth. Composer 1 initially just gave me false information (I had read the docs myself, so I knew it was fabricating) about Better Auth’s capabilities.

This is just unacceptable. The @Docs feature and doc indexing is one of the reasons I signed up for my Ultra plan in the first place. I started using this very early on, back around April time frame this year, and the improvement in the quality and correctness of code implemented by the models with @Docs was SO FAR BEYOND normal usage, that @Docs usually found its way into nearly all of my prompts.

Now the models cannot use the @Docs feature, and the quality of the code produced is pitiful in comparison, the models are constantly guessing and assuming. They will guess and assume even after extensive web searches, so web search is NOT a viable alternative to @Docs.

I’ve been asking you guys what the deal is with @Docs for months now. I really would like to know: Is this a feature you guys are going to fix? OR are you just going to ignore it and let it die? There is no alternative for @Docs, as far as I have ever been able to find. This was one of the SINGLE MOST POWERFUL features you guys had, and you killed it months ago and seem fit to let it die.

I really, heavily, persistently RELIED on this feature, and I am really starting to have a hard time using Cursor without it. My current use case, the darn models just cannot seem to fully understand Better Auth and they are just giving me results that seem to come from rudimentary understanding from their latest trainings. Not the deep, ultra rich and EXACT knowledge they used to have when @Docs worked.

Sonnet 4.5

Composer 1

The ridiculousness here, is Composer 1 seemed to be able to include a reference to @Better Auth Docs in its response, figured the docs were “indexed elsewhere”, but couldn’t access them!

I do wonder if this is part of what appears to be a deeper issue with MCP support in Cursor in general. I’ve pushed the models to use @Docs hard in the past, and they will usually eventually try to connect to it as an MCP server. I assume that is, in fact, how its implemented, that it is basically an MCP and exposes tools? Is there a bug in how the @Docs context and tooling is exposed that was introduced back at the end of August or beginning of September?

I also wonder if there may be some Cursor Account level limitation here. This issue seems to FOLLOW ME AROUND, from one install to the next, even the most pristine install possible. Is there something misconfigured, corrupted, or otherwise broken with my Cursor account, that is preventing this feature from working?

I really need this feature. Second only to the terminal issues that plagued Cursor for several months before 1.7.35 was released, lack of @Docs is the single most impactful bug that affects my work right now with Cursor. It has a huge impact on the quality and correctness of the models, in terms of implemented code, but also just understanding how any given third party library, framework, CLI, or other tooling works, which then leads into planning, software design and architecture, etc.

Steps to Reproduce

Index one or more third party library docs sites (in my case, Better Auth).
Ask the model to use the docs via the @Docs command palette.
Weep at the lack of actual reference via @Docs, and subsequent misinformation, obfuscation, guesswork, assumption, and fabrication of false knowledge, bad implementations, etc.

Expected Behavior

@Docs should work, the way it did before, providing deep, rich insight into how third party libraries or common frameworks work, without the need for web searches, guesswork, assumption or fabrication by the models.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.0.64
VSCode Version: 1.99.3
Commit: 25412918da7e74b2686b25d62da1f01cfcd27680
Date: 2025-11-06T04:35:14.424Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 25.0.0

For AI issues: which model did you use?

Composer 1
Sonnet 4.5

For AI issues: add Request ID with privacy disabled

fc110e51-e094-49dc-9de6-0c6d3502fd63 (Composer 1)
42296efa-909e-43b2-af16-0a1608c0f08b (Sonnet 4.5)

Does this stop you from using Cursor

No - Cursor works, but with this issue

3 Likes

@condor @andrewh @deanrie

Wow.. Your post editor seems to have some severe bugs as well. Editing my post seems to have totally botched most of it… Could you guys fix that as well?

please fix this already. @docs haven’t worked for weeks now!

2 Likes

Well, Docs just seems to keep degrading. They don’t even list in the command palette anymore. There used to be both any custom indexed docs, as well as a bunch of “standard” or “preindexed” docs for a whole bunch of things. Now it shows nothing:

Guys, what’s the plan here? Are you going to just remove the @Docs feature? Or are you going to fix it? I’d really like to know… This limbo state is infuriating.

Version: 2.0.69
VSCode Version: 1.99.3
Commit: 63fcac100bd5d5749f2a98aa47d65f6eca61db30
Date: 2025-11-07T18:21:29.650Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 25.0.0

1 Like

GMP to all. Antonio Rivas here from Toluca, MX.
—
Version: 2.0.69 (system setup)
VSCode Version: 1.99.3
Commit: 63fcac100bd5d5749f2a98aa47d65f6eca61db30
Date: 2025-11-07T18:21:29.650Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26100
—

I think Docs is powerful feature, at the beginning I was able to index several large documents using a local server with python and a certain port. So would input something like http://127.0.0.1:8081/ and the docs feature would index the whole document to be referenced in Cursor chat. But I’m suspicious that this would put a massive load in Cursor Servers, which would explain the hesitance from developers from further offering a most robust “Docs” system. So, I rolled back to Google NoteBookLM which is currently free to try (test it before they charge⚠️) which from my opinion is a well made PDF ChatBot (although many other formats it can take) then I reference the outputs to Cursor AI.

For what I’ve been reading, Cursor AI is exploding in customer and complexity, which isn’t that nice for the staff, when your puppy :dog_face: is monstruos, LOL.

Docs still taking websites with minor index items that you can see when you click the preview on the doc but now refusing to take localhost docs. I’m seriously thinking making my own local API for docs embeddings, perhaps Ollama? anyone?

Grace from JC to us.

I also miss the feature, hope they bring it back.

Thanks for the detailed reports! This is a known issue that’s being tracked.

To help add your case to the investigation, could you verify a few things:

  1. Cursor Settings > Indexing & Docs - Are your Better Auth docs showing as indexed with a green checkmark?
  2. Try in a fresh chat - Does @Better Auth Docs appear when you type @ in a brand new chat?
  3. Request IDs - The ones you provided are helpful. If you test again in a fresh chat, please share that request ID too.

The team is investigating this. Another related report: Can't go through docs

Hi Sanjeed,

Glad to hear you guys are tracking this. Regarding your questions:

  1. Yes, all docs (except a couple very large ones, key example RabbitMQ) show green checkmarks.
  2. This issue has been going on for months. Literally. Not only have I tried with new chats, I’ve tried with completely clean installs of Cursor, as well as fresh installs on BRAND NEW MACBOOKS, and also on my Windows computer as well. The issue is persietent, across all of these factors. This is one of the reasons I wonder if there is an ACCOUNT LEVEL factor here…maybe it is not just Cursor itself, but something screwed up with my Cursor user account? Some of my coworkers have similar issues with @Docs, others seem to be able to use @Docs fine. So something is very strange here, and I think you guys should consider account-level factors, as well as maybe regional factors (i.e. geographic region issues with servers that support @Docs?) I suspect @Docs is in effect an MCP, so if the model has to call back into it, maybe there could be regional issues there…
  3. I haven’t bothered using @Docs in a while, and with this brand new computer, I have only a handful of docs indexed now. I can give it a try, but I don’t think you guys will glean anything more than from the previous request ids I shared. The issue has been not only persistent but consistent, across all the various installs mentioned above.

Glad to hear you guys are finally investigating this, though. This is one of the KEY features that hooked me on Cursor in the first place. If it is left broken or removed, its a MAJOR reason I would move to a different platform for agentic coding, as it is, IMO, one of the single best ways to get any model to implement the best quality code possible. Not having this feature the last few months has been extremely frustrating and has had an immense impact on the quality of the code the agent creates.

Also, I don’t know if this is a feature that only Claude models support or not. If so, it would be great to see support for @Docs expanded beyond Claude…to Grok Code, Composer 1, at the very least.

Cursor is essentially useless without this feature. What is the plan in terms of refunds on subscriptions for the time period that this bug is still active?

1 Like

@sanjeed5 Is there any update on this? Are you guys still working on fixing the docs feature?