Studio MCP Server Updates and External LLM Support for Assistant

Updates Announcements
, ,
211

Different MCP clients approach approvals and confirmations flows differently and can change with updates, so we suggest reading up on what settings your client of choice may have.

For Codex, this resource maybe helpful:

212

I am aware of this, but it’s only been in the last couple of days that Codex has decided to ask for permissions at certain points when editing scripts and yet not at others.

I’m not sure why some script edits work and some don’t but I’m not entirely sure how Codex is responsible for deciding which Studio actions require permissions.

That said it’s already doing it a lot less today and Claude never does so it’s no big deal.

213

I accidentally asked it something during a playtest, and now in studio it created folders while looking for something that only exists on run-time:

Yeah that’s a nasty gotcha — the MCP creates parent folders/instances along the path when the target doesn’t exist. I’ll always use the edit-mode path (ServerScriptService.WeaponHandler.Utils.CooldownWidget) going forward, never the runtime path.

214

image
image290×128 4.5 KB

Like this makes no sense?? it creates folders when it goes down paths that don’t exist

1 Like
215

I’ve been using the MCP for a bit and it’s super super super good overall however I think some things could be added which would significantly improve it

  1. Adding a switch to server tool
    Example: get_current_context, switch_context(server | player) and allowing execute_luau to run on the server

This would make debugging significantly easier, as it would allow the agent to switch between the server and client during a playtest. For example, imagine you’re testing play-time rewards and you don’t want to wait through the full timer, so you temporarily reduce the requirement from 10 minutes to 5 seconds just for testing. Even then, you’ll eventually have to go back and change it. Adding the ability to switch context would allow you to call the reward function directly without having to change any values (speeding up your workflow).

Another example: imagine you’re working on a game in which monsters spawn randomly, but one monster has a bug where it doesn’t spawn. Right now, the agent has to keep testing and playing until that monster spawns. With the ability to switch context and run code, the agent could simply spawn the monster directly, skipping the whole waiting process and getting to the bug and fixing it

BUT this does have risks like touching DataStores, so maybe a safer option would be a tool that allows the agent to call a function that exists on the server, but only if it’s whitelisted by the developer, similar to a Roblox config file that the agent cannot modify and strictly follows.

217

Could you please allow Code modification during playtests? By e.g. separating the studio or so
It would make things much smoother, as of now I have to start separate testing sessions via Player1 etc if I want to test while it’s writing code

218

image
image786×138 6.42 KB

Another bug, it can’t access some things

219

Would also appreciate if the Player1, Player2 etc test could be 10x faster (my hardware should allow it, just an optimization of Studio stuff)

This is the way I test rn while mcping code changes