Best way to reference multipage API/SDK Docs?

Discussion
1

This has been discussed, but I couldn’t find the answer.

What is currently the best way to reference API/SDK docs in your chat, so that the agent will properly follow those rules? I find multi page docs difficult when trying to link a URL, because it’s not going to crawl the index.

2

Have you tried using the built-in documentation index system?

Cursor Settings > Features > Docs

Add the index page for the documentation you want; Cursor will index it. Then, you can @mention it inside your chats, and your agent will simply pull what it needs from the pages it needs.

1 Like
3

how about for a vercel based website where the index isn’t clear. Example the docs index system only retrieves 1 page from this example:

4

They are using an iFrame which I doubt the indexer will pickup on.

Try this URL instead.

5

Same deal, one page result shown. It’s better than nothing, but difficult to get the granular information for classes, funcs, structs ect.

I could manually extract it all I suppose since it wouldn’t take me forever but was seeing if there’s another way.

6

I messaged in the discord to change the formatting of the docs. I think a business will be created around this, or a standard will emerge for reference guides and docs to be agent compatible. It’s a HUGE unlock for coding.

7

Yeah, the root issue is how their documentation is set up. The page that Cursor would look at can’t see the index on the left, so it only indexes the one page.

I tried digging through the front-end code to see if that sidebar can be pulled in, but it doesn’t appear so.

You might be able to try some of the other tools that can index documentation and turn it into Markdown to use locally, which Cursor could query.

The link I gave, though, will provide you with the direct link to each of the pages you need, which should make @mentioning them inside of chat easier.

Then you’re giving a URL like this: https://docs-ios.pipecat.ai/PipecatClientIOS/documentation/pipecatclientios/boterror/init(from:)

Instead of it always being: https://docs-ios.pipecat.ai/

It’s not ideal, but it at least gets you closer.

Another option is to add Brave Search MCP. Then you can ask it to find the info it needs. You’ll need a Brave Search API account, but they allow a query per second and 2,000 credits a month for free. I’ve yet to exceed it.