Is there a document function like cursor that can index some documents as a knowledge base?

[dnslin]

That way, you don't have to crawl the site every time, which reduces token consumption.

[kwanLeeFrmVi]

roo don't have it now, so I make it for my own, but happy to share it with everyone.
kwanLeeFrmVi/mcp-rag-server
hope they can integrate it into roo. that will help a lot

[tapestreamer]

From https://www.reddit.com/r/RooCode/s/H4OnZxIVMq

Feature Request: Cursor @docs... a must have for coding reliably

One critical feature preventing me from switching to RooCode is the lack of a robust documentation pre-population system.

I've been coding for over 20 years and I use AI coding tools extensively... so please here me out before you suggest some alternative.

"Storybook is constantly adding new features and deprecating stuff. You sort of always need to reference their documentation when coding for the most reliable results.")

When working with AI coding assistants, the single most effective way to improve code quality and accuracy is feeding version-specific documentation about libraries and systems directly into the AI.

Why Runtime Documentation Retrieval Isn't Enough

Current approaches to documentation handling (grabbing docs at runtime via MCP Server or specifying links while coding) fall short for several critical reasons:

1. Version specificity is crucial - Example: asdf-vm.com has completely different instructions for v16+ versus older versions. In my extensive experience, AI consistently defaults to older (albeit more widely used) documentation versions.
2. Performance impact - Retrieving and indexing documentation at runtime is significantly slower than having it pre-populated.
3. Reliability and accuracy - AI frequently retrieves incorrect documentation or even hallucinates functionality that doesn't exist in libraries/frameworks. Pre-populating documentation eliminates the frustrating "no, here's the correct documentation" dance I regularly experience with AI assistants.
4. Context switching kills productivity - Maintaining separate documentation links and manually feeding them to AI during coding sessions creates unnecessary friction. Suggestions to "process my own documentation, create markdown files, and then feed them into the system myself" only add more overhead to my workflow.

The Solution: Cursor's '@docs' Implementation

https://docs.cursor.com/context/@-symbols/@-docs

Cursor's implementation prevents me from using any other AI editor because it provides:

• Pre-indexing capability - I can enter a website URL, and Cursor will scrape and index that information for reference in subsequent chats
• One-click refreshing - I can simply hit refresh in the documentation panel to re-index any site for up-to-date documentation

"All my documentation indexed in one place in cursor, with a custom label, the date and time it was indexed, whether the indexing passed or failed, and the ability to refresh the index to pull the latest up to date documentation, and to even see the pages it indexed. No other AI tool has this.")

• Flexibility - I can use ANY URL as documentation, whether it's official docs, GitHub pages, or specialized resources I personally prefer
• Seamless workflow - I can stay inside the editor without using external tools, managing documentation links, or creating custom setups

This feature dramatically improves code quality to the point where any AI coding editor without this capability is significantly handicapped in comparison.

Why This Matters for RooCode

If RooCode wants to compete in the AI coding assistant space, this isn't an optional nice-to-have - it's a fundamental requirement for serious developers working with complex, version-dependent libraries and frameworks.

For professional developers like myself who rely on AI assistance daily, the ability to pre-populate specific documentation is the difference between an AI tool that occasionally helps and one that becomes an indispensable part of my workflow.

[cvladan]

I know it's not exactly the same, but there are similarities, so please support Markdown in the codebase indexing functionality.