ZeroScript is a free browser extension that turns ChatGPT, DeepSeek, Gemini, Kimi, GLM, Qwen, Arena or Meta AI into a Roblox Studio AI agent. Control Roblox Studio with AI directly from your browser - read/edit scripts, run Luau, generate assets, all from a normal AI chat. No API key, no terminal, no coding needed.
🌐 Website: zerodev.tools/zeroscript the free Lemonade.gg / Luamotion alternative for building Roblox games with AI.
Eight AI providers are supported: DeepSeek (chat.deepseek.com, recommended), ChatGPT (chatgpt.com), Google Gemini (gemini.google.com), Kimi (kimi.ai, Moonshot AI), GLM (chat.z.ai, Z.ai), Qwen (chat.qwen.ai), Arena (arena.ai, a multi-model playground) and Meta AI (meta.ai). On ChatGPT, screenshots and image input are turned off on purpose: the free tier limits files and images on a separate quota from messages, so vision would only work part of the day. Gemini and Kimi can be unstable: Gemini tends to stop using the Roblox tools in long sessions, and Kimi sometimes uses its own native tools instead of the Roblox commands. On Arena, use Direct mode (ZeroScript only supports Direct; it blocks Start in Battle / Side-by-Side / Agent modes). DeepSeek is the recommended provider.
💬 Stuck? Join the Discord community get help, share feedback, and follow updates.
Also known as: ZeroScript Roblox, ZeroScript free download, Roblox ChatGPT agent, Roblox DeepSeek agent, Roblox Gemini agent, Roblox Kimi agent, Roblox GLM agent, Roblox Qwen agent, Roblox Arena agent, Roblox Meta AI agent, Roblox Studio AI automation, Luau AI, MCP Roblox, lemonade alternative free, lemonade.gg alternative, free Roblox AI agent, free lemonade roblox alternative
ZeroScript is 100% free and open-source. It always has been, and it always will be. There is no official paid version, no subscription, and no sign-in required to use the extension.
If you come across a site or extension using the ZeroScript name that asks for payment or account creation, it is not this project. The only official links are the ones listed at the top of this README.
AI chat (ChatGPT / DeepSeek / Gemini / Kimi / GLM / Qwen / Arena / Meta AI, in your browser) -> ZeroScript Extension -> Bridge (your PC) -> Roblox Studio
The extension runs inside the chat page (ChatGPT, DeepSeek, Gemini, Kimi, GLM, Qwen, Arena or Meta AI). When you type a request, it sends commands to the Bridge running on your PC, which drives Roblox Studio through the built-in MCP server.
📺 Lost? Watch the setup tutorial on YouTube it covers every step below.
Download the latest zip from the Releases page and extract it. The zip contains both the Bridge and the extension folder.
To load the extension:
- Go to
edge://extensions(Edge) orchrome://extensions(Chrome) - Enable Developer mode (top right toggle)
- Click Load unpacked
- Select the
zeroscript-extensionfolder from the extracted zip
Open Studio and load a Place, then enable MCP (first time only):
- Click Assistant AI in the top bar
- Click ... (top right of the Assistant panel)
- Click Manage MCP Servers
- Click Enable Studio as MCP Server
Not sure where to find these options? The video tutorial shows exactly where to click.
- Windows: double-click
start.batinside the extracted folder. - macOS: double-click
MacOS_Start.commandinside the extracted folder. The first time, macOS will show a security warning ("could not verify... free of malware") - this is normal for any script downloaded outside the App Store, click Done, then go to System Settings > Privacy & Security, scroll to the bottom, and click Open Anyway. You only need to do this once.
A small window opens, that means the Bridge is running.
Go to https://chat.deepseek.com (recommended), https://chatgpt.com, https://gemini.google.com, https://www.kimi.ai, https://chat.z.ai, https://chat.qwen.ai, https://arena.ai or https://www.meta.ai and open a new chat. The ZeroScript bar appears above the input box. Click Start session. Type what you want to build.
Only works on chat.deepseek.com, chatgpt.com, gemini.google.com, kimi.ai, chat.z.ai, chat.qwen.ai, arena.ai and meta.ai - it will not work on any other site. On Arena, keep the mode dropdown on Direct - ZeroScript blocks Start in Battle / Side-by-Side / Agent modes (it only drives a single Direct reply). Gemini and Kimi can be unstable (model behavior, not the extension): Gemini may stop using the Roblox tools after a while, and Kimi may use its own native tools instead. If the AI starts answering in plain text instead of acting, remind it to use the commands or start a new session.
Watch the setup tutorial on YouTube
- Read and edit scripts
- Run Luau code directly in Studio
- Inspect the game tree and instances
- Generate meshes, materials, and models
- Browse and insert from the Creator Store
- Control play-testing
- Remember your project across sessions persistent project memory saved inside your place
- Kimi moved to kimi.ai. The old address, kimi.com, now asks for a Chinese phone number to sign in, which locked most people out. Open https://www.kimi.ai instead - the page is unchanged, the bar appears above the input box exactly as before. Reopen any Kimi tab you had on the old address.
- DeepSeek: the Instant model can now run the agent. Picking Instant used to leave "Start Roblox agent" spinning forever with no explanation, because only Expert and Vision were accepted. Choose Instant before starting and the session runs on it - much faster than Expert, without the reasoning pass. Images stay off on Instant just like on Expert; the Vision tab remains the only one that can see screenshots.
- DeepSeek: a reply written in DeepSeek's own tool-call format no longer kills the turn. DeepSeek occasionally answers with its internal markup instead of a ZeroScript command. Nothing recognised it, so the tool never ran, the raw tags stayed on screen and the agent stopped dead with you waiting. It is now caught, hidden behind a tool chip like any other command, and DeepSeek is told to rewrite the call properly.
- ChatGPT: the bar no longer clips into the composer's rounded corners.
- ChatGPT: you can chat normally again without starting an agent. On a blank ChatGPT tab the extension refused to let a message send until you clicked "Start Roblox agent" every other provider only suggests it, ChatGPT was the odd one out.
- ChatGPT: it no longer forgets it can actually run commands. ChatGPT summarises its own context mid-session and the first thing it drops is the mechanism it then tells you "I can't invoke those commands in this session" while the extension sits there, ready. Its instructions are now re-stated automatically, carried along on a tool result so they cost no extra message and stay hidden from you (a "Reminder" chip marks them). Tool results are never shortened to make room.
- ChatGPT: an image you send is now used as reference for the work, instead of being answered with a freshly generated picture. Ask explicitly if you do want an image.
- A finished command is no longer stranded as "not run" after a long reply (seen on Qwen writing for 400s and more), where the agent used to give up eight seconds after the model stopped.
- A clear message when ZeroScript updates while a tab is open. This used to be reported as "the bridge stopped on your PC run start.bat", sending you to fix something that was never broken. It now tells you the page needs reloading and offers a Reload button.
- The AI no longer insists your bridge is offline without checking it must run a command first before saying so.
- ChatGPT support (chatgpt.com) an eighth provider. Screenshots and image input are off there on purpose: ChatGPT's free tier limits files and images on a separate quota from messages, so vision would only work part of the day. The model picker and reasoning mode stay entirely your choice.
- ChatGPT: fixed most tool calls failing. ChatGPT renders code blocks with an editor that keeps no line breaks in the page, so a perfectly valid command was read as one giant line and came back as "your code block was empty". Replies are now read with their real line structure.
- ChatGPT: fixed long commands running truncated. Past roughly 2000-4000 characters the page only shows part of a long line and then stops updating, so a big command executed cut off (the tool chip's token count would climb, fall back to ~500, and freeze). ZeroScript now reads the editor's true content instead of what's drawn on screen a 5.3k-token
multi_editapplies whole. - ChatGPT: fixed the raw command text staying visible when the model wrote it outside a code block.
- Meta AI: fixed big commands failing with "bad JSON". Meta shows a JSON block in an interactive viewer that shortens large values - a 19k-character
multi_editappeared in the page as 223 characters ending in"edits":[1 item], so the command was read truncated and rejected. This is also what made the tool chip's token counter collapse to ~44 tokens when the block finished rendering. Commands are now read from the viewer's Raw tab, in full. - Clearer message when Roblox refuses to parse your Luau. ZeroScript used to always blame an empty code block or a wrong
###LUA###marker, even when a full script had been sent - so the model "fixed" something that wasn't broken and failed again. It now says how many characters were sent and names the real causes: invalid syntax, or code too large for the parser. - Fixed a stylesheet error that silently disabled command hiding on GLM, Kimi, Qwen, Arena and Meta AI.
See CHANGELOG.md for older releases.
| Dot | Meaning |
|---|---|
| Green | Bridge + Studio ready (a place is open) |
| Yellow | Bridge OK, but Studio isn't usable yet - open Roblox Studio, load a place, or enable its MCP server (hover the dot for the exact reason) |
| Grey | Bridge offline - run start.bat (Windows) or MacOS_Start.command (macOS) |
- Windows or macOS
- Roblox Studio (MCP support built-in)
- Microsoft Edge or Chrome
- Python 3.9+ (installed automatically on Windows, or install it yourself on macOS - see python.org/downloads)
ZeroScript is free. If it saves you time: Ko-fi - Robux tip passes available in the extension panel
Credit: the idea for connecting other MCP servers (Blender, Sketchfab, etc.) alongside Roblox Studio came from javnpa.
Credit: macOS/Linux support contributed by archivealf.