Best way to integrate/add support for proprietary language

We have our own language, where there’s not a lot of publicly available source code examples, and as such foundation models wouldn’t have it in their training data.

We would like users to be able to use Cursor in order to write applications using our language. What’s the best/smoothest way to enable such an integration?

We do have some finetuned OpenAI Azure models, and could do the same with other providers. We’re also experimenting with few-shotting (just attaching a few good examples to the context), which seems to be promising.

MCP only seems to support tool usage in Cursor right now, so I guess MCP sampling is not an option. Are there any other good approaches to curating relevant context construction in a way that’s easy to integrate in Cursor for users?

There are a few good approaches you can take to improve Cursor’s support for your language:

  1. Add your language docs using @Docs - this lets you crawl and index custom documentation for context

  2. Set up project rules in .cursor/rules to define language-specific patterns and preferences. Check out the rules docs for more details

  3. For the few-shot examples, you can use @Code to reference specific code snippets as context. This way users can easily access example code while working

  4. Enable codebase indexing to help Cursor understand file relationships and code structure

The finetuned Azure models won’t work directly with Cursor right now since requests go through Cursor’s infrastructure first. Your best bet is probably focusing on good documentation integration and example code context

Let me know if you want me to clarify anything about setting up the docs or rules

Thanks! That is indeed what I’ve done personally when trying things out, but I was more wondering if there are any good approaches to offering the same sort of experience to our users in a low-friction way.

For a low-friction user experience, you can create a shared rules file that users can drop into their .cursor/rules folder to instantly set up the language config.

Another option is to create a template project with all the context and config pre-setup that users can clone to get started quickly, if that would be helpful!