RAG-like process (for Cursor) on local auto generated documentation ?

[janot51]

Hi all,
I managed to reverse engineer several software projects (with no documentation) and to create a bunch of *.md files and then mkdocs wikis.
Question is simple : is it possible to help Cursor/LLM prompt to get the context of this documentation in order to be able to answer quick and with more precision : RAG-like process?
Today I just add the doc folder to the context with @doc. It wotks but Cursor need generally to open multiple files to get the needed information.
Thx

[jlcases]

Hi Janot,

We use the repo-pack library to build context from code before generating documentation. I believe you can also create single-file documentation with repo-pack and use it to query a specific file from Cursor.

You can do this using the GENERATE_CONTEXT command with PaellaDoc.

That’s my recommendation based on our experience

[janot51]

Hello,
I tried this:
I converted my *.md files into a single .txt file using repo-pack.
I then uploaded this text file to a GroundX RAG platform bucket.
I installed an MCP server in Cursor to link to my GroundX RAG bucket.
Now, in the Cursor prompt, I tell the LLM to try to find information with a RAG query.
Based on the result, the LLM quickly has good context and can optionally open specific local files to refine the search.
This seems to work well.
Perhaps a bit too complicated compared to Cursor's indexing capabilities?
The advantage of this solution: I can add other documents (e.g., PDFs) to my GroundX bucket. Cursor can then index them automatically.