# Using Cursor and llms.txt

**URL:** <https://discuss.ocaml.org/t/using-cursor-and-llms-txt/16545>\
**Category:** Learning\
**Created:** [April 24, 2025, 12:25pm UTC](https://discuss.ocaml.org/t/using-cursor-and-llms-txt/16545 "2025-04-24T12:25:11Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![nojaf](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ocaml.org/nojaf/32/5587_2.png) [@nojaf](https://discuss.ocaml.org/u/nojaf)\
**Post date:** [April 24, 2025, 12:25pm UTC](https://discuss.ocaml.org/t/using-cursor-and-llms-txt/16545/1 "2025-04-24T12:25:11Z")

</div>

I start using cursor, and I wonder which urls people use to add additional context for the OCaml language.

You can add documentation via urls:

![image](https://us1.discourse-cdn.com/flex020/uploads/ocaml/original/2X/0/0720259787370ed637fea7c4e60fc5015da0a33b.png)

I’m wondering if anyone has any good resources to add here.  
At first glance, the [Learn OCaml](https://ocaml.org/docs) doesn’t provide a [llms.txt](https://llmstxt.org/).  
Does anyone have some alternatives?

---

<div class="post-metadata">

**Author:** ![avsm](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ocaml.org/avsm/32/6_2.png) [@avsm](https://discuss.ocaml.org/u/avsm)\
**Post date:** [April 25, 2025, 6:39am UTC](https://discuss.ocaml.org/t/using-cursor-and-llms-txt/16545/2 "2025-04-25T06:39:25Z")

</div>

I’ve got an MCP OCaml implementation [here](https://tangled.sh/@anil.recoil.org/ocaml-mcp) you can “try at your own peril”; that can be used to write a local odoc query for Cursor.

This is _entirely_ agentically coded as I’m experimenting with models, but MCP seemed an appropriate protocol to try that out for! However, it means the quality/risk of using it is definitely “buyer beware”.

---

<div class="post-metadata">

**Author:** ![Khady](https://sea2.discourse-cdn.com/flex020/user_avatar/discuss.ocaml.org/khady/32/469_2.png) [@Khady](https://discuss.ocaml.org/u/Khady)\
**Post date:** [April 26, 2025, 3:42am UTC](https://discuss.ocaml.org/t/using-cursor-and-llms-txt/16545/3 "2025-04-26T03:42:01Z")

</div>

Context is probably a bit different, because ahrefs is using a large monorepo and tradeoff are different, but we’ve been experimenting with writing pretty high level introduction prompt per technology. For example we have one for melange. I used openai deep research and Claude to write the initial version. Then we are starting a library of prompts related to specific libs/patterns/refactoring. Some are systematically used, some are loaded for given path, some are loaded automatically by the agent when it deems it useful.

Then for links to docs I actually point as generic documentation rather than llm specific text. For example the api of lwt, the one of melange, the OCaml manual, the page about runtime from RWO, …

It’s hard to evaluate the impact of each of those. The outcome seems to vary a lot depending on the model and the prompt and the task at hand. But I feel like it would be worth putting some efforts at least as part of an experiment to have comprehensive prompts a la cloudflare. See [Prompting · Cloudflare Workers docs](https://developers.cloudflare.com/workers/get-started/prompting/)