MCP Feature Request: generate output in project directory

[gabimoncha]

I've somehow managed to chain different repomix mcp tools together, but anytime i try to access it, i get this Cursor error: Content type "resource" not supported

would be awesome if pack_codebase could have a local parameter so it woud generate the repomix output in the project directory

@yamadashy wdyt?

[yamadashy]

Hi, @gabimoncha !
Thanks for the report!

Ah, I see—so returning a resource from a tool doesn't work well with Cursor.

The local option sounds like a good idea!

I'll think about it a bit and see how it might work.

[gabimoncha]

Thanks @yamadashy. I saw the PR and tried the new version, but it still doesn't work.
Maybe I'm no using it correctly?

For context, I'm trying to integrate repomix in my cursor-rules-cli library, to help users generate better AI instructions.
https://github.com/gabimoncha/cursor-rules-cli

[yamadashy]

@gabimoncha

Sorry, this issue hasn’t been fully resolved yet — I only fixed the part that returns the resource.

Could you try giving an instruction like “Pack with repomix in the current folder”?
The packing tool and the reading tool run separately, and the outputId is only available after packing.

You can also provide a GitHub URL.

I’ll work on the local option you suggested separately.

[gabimoncha]

thanks @yamadashy. but that one doesn't help me, since my CLI packs repomix with some custom instructions for the Cursor agent. let me know if I can assist you with anything

[yamadashy]

If customization is needed, it might be better to call repomix directly instead of going through MCP.

Currently, the main use case for MCP is to pack a local folder or remote repository that you want to reference and apply it to the current chat.

If you have a need to pack the current folder and use it that way, I would be happy to improve it in that direction, including the local option you suggested.

[GitMurf]

So that I am clear on the expectation of how this works, does this mean there is not actually any output file when packed by the mcp? is instead it just piping directly to stdio or something?

Secondly, it sounds like from your comments @yamadashy that the repomix mcp is more used for code outside your project like remote repositories or another folder on your filesystem or maybe a subdirectory in your project? Is this accurate? And with that, is the idea that if you want to actually use repomix in an LLM conversation on your codebase, you can/should just run repomix (not the mcp server) so that it creates the output xml and then you can just use/reference that output xml for your codebase in your conversation with the LLM as opposed to having the MCP create it for you?

Am I understanding generally the intended purpose of the mcp? I am new to this project so please don't hesitate to correct me if I am mistaken in any way :-)

Thanks for the great work!!