Cursor setup: .mdc rules, agents.md, and hooks — what actually works?

[oliviacraft]

A common question in the Cursor community: how do you structure .mdc files, agents.md, and hooks when you have multiple languages in the same repo?

Here's what works:

For multi-language repos (Python + Java + TypeScript):

Three separate .mdc files in .cursor/rules/:

python-conventions.mdc   (glob: **/*.py)
java-conventions.mdc     (glob: **/*.java)
typescript-conventions.mdc (glob: **/*.ts, **/*.tsx)

Set to Auto Attached — only loads when matched files are in context. Token-efficient and precise.

agents.md vs .mdc: agents.md in a subdirectory acts like a scoped Always Apply rule for that folder. Works, but a named .mdc with explicit globs is more maintainable.

Hooks vs skills: Hooks fire automatically on tool events (after Edit, after Bash). Skills are invoked on demand. Use hooks for always-on side effects, skills for optional workflows.

Full setup guide as a Gist: https://gist.github.com/oliviacraft/281f8ba971d238e8ff0ae87e2e99f21c

Free starter kit with pre-built rules for 14 stacks: https://oliviacraftlat.gumroad.com/l/pomoo
Full pack ($27): https://oliviacraftlat.gumroad.com/l/wyaeil

What's your setup?