Will extensions be able to access the cursor ai features via an API?

Feature Requests
1

Would be great if my extension could talk to cursor’s ai feature. Current idea is to have my extension provide more context and dynamic docs/data to the cursor embeddings so the chat can give better answers

3 Likes
2

That’s an interesting idea. Of course, Cursor could implement your extension idea themselves but this could lead to some interesting extensions. I would imagine that they might be a little hesitant to add this sort of API in general because it could lead to extensions taking up a lot of resources on their GPT key. But with the right implementation, (and maybe an additional fee for using this API) I’d like to see what people could come up with.

3

Love this idea! If people have concrete suggestions for the API interface (i.e. what info would you want to ingest, what actions would you want to perform), that’d be great.

We want Cursor to be veryextensible.

4

A good start is to expose the existing APIs you use to allow users to add docs and set rules/prompts as APIs extension developers. I get there are safety and abuse concerns here but there’s opportunity for the community to come together and create really awesome niche use cases so the cursor team doesn’t have to!

2 Likes
5

Is this possible today? I am okay paying extra for this feature, basically, API access to cursor features so I can connect it to a subsystem within the organization.

6

It is indeed a great idea. I have Postman plugin installed. If the request/response can be referenced as read-only docs, that’d be very cool.

7

My ideal workflow would be if i could create a semantic RAG vector db of my entire codebase, and then have an extension or api to the chat/compose tools in cursor, so my db would be searched for context to include before the llm query is sent out.

Maybe the MCP protocol is relevant and could be used here?

Good point above about the impact to token usage for Cursor. Maybe this extension could only work if users were using their own llm keys?

8

My workflow would be something like this:

Create tasks of new features on my personal application. My automation workflow would see there’s a new task and “ask” Cursor to build it based on some documentation. Cursor builds and commits, then my pipeline takes care of the rest.

When will something like this be possible?