Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
I am a prolific docs user. I reference docs CONSTANTLY. However, in the latest versions of Cursor, they do not seem to work. I didn’t realize it at first, but, it seems that NOTHING can use the @DocsDocsDocsDocs feature now. It used to be that Claude was the only one that REALLY seemed to use it. It doesn’t seem to be able to use it AT ALL anymore, and instead, I have now noticed it falls back to doing web searches. Web search is not reliable, and not even remotely as good as directly referencing known-good, explicitly indexed documentation web sites EXPLICITLY for the SPECIFIC libraries I need the agent to fully, and completely understand.
This has had a devastating impact on the ability of the model to actually do the job optimally and very quickly. I was just working with class validator, and it seemed to be spinning its wheels DESPITE giving it docs references. I then told it to read the docs, and it searched the web. I told it NOT to search the web, and it searched my code. I then explicitly referenced the docs again, told it to use the explicit docs reference I supplied…and it simply couldn’t seem@Docshet@Docs use @Docshe @Docs feature.
I have been using Grok Code a lto, and used GPT-5@Docsfor a while, and @Docs did not seem to work for either of them. Recently, I’ve been using a mix of Grok Code and Sonnet. At first, which I think I may have still been on Cursor 1.4 or 1.5, it seemed like Claude was able to use the docs feature properly, however in 1.6 (maybe later 1.5 versions) I am now noticing it DOES NOT, and instead uses web search (whcih is really hit or miss, and even on a hit, it just is not even remotely as good as REAL pre-indexed docs usage.)
This is one of the most significant features for me, one of the key reasons I have stuck with cursor. The, possibly PRIOR, docs ingegration, was extremely powerful, and the difference between using docs, and not, was HUGE in many cases. This is one of the key reasons I still use Claude, it seems to be the only one (or at least, WAS the only one!!!) that could actually use the feature, and it has such a dramatic impact to the quality of results.
Loss of this feature is devastating. Not only does it need to be restored, it REALLY needs to be expanded to other models. Grok Code and GPT both need to be able to use preindexed docs the same way Cursor did before.
EXAMPLES:
Said it could “see” the docs, but simply couldn’t use them? Even tried to list mcp resources, no resources found? (I actually have several other mcps configured, even…) Couldn’t find or use or even seemed to know how to use the docs feature. Then just went to checking the codebase. The darn thing, has been spinning its wheels, trying to use class-validator decorators properly. It seems to just be GUESSING at how to use them, and iterates and iterates over and over trying to get each usage correct, and its always off a little. With @Docs in the past, this kind of wasteful guesswork would have NEVER occurred. It would have known exactly how to use class-validator, and used it properly the first time. Or known exactly when it needed to create a custom validator.
In this case, it searched the web, instead of using the docs, which are indexed right there in Cursor:
Index:
Steps to Reproduce
Reference soem indexed docs in a prompt.
Weep as Claude relies on web search instead.
Expected Behavior
DOCS ARE ACTUALLY USED!!
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
Version: 1.6.27
VSCode Version: 1.99.3
Commit: d750e54bba5cffada6d7b3d18e5688ba5e944ad0
Date: 2025-09-17T20:21:17.042Z
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.5.0
For AI issues: which model did you use?
claude-4-sonnet (thinking)
claude-4-sonnet
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor
