AI Assistant for Roblox Studio

Multi-AI assistant that lives inside Roblox Studio. Chat with AI directly in Studio to read, write, and modify your project. Supports multiple free AI providers — no paid API keys required.

Supported AI Providers (All Free)

How It Works

You type in Studio plugin chat
        ↓
Plugin sends message to server (HTTP POST /chat)
        ↓
Server calls AI provider (Groq/Gemini/Ollama/HuggingFace/Claude/Mistral/DeepSeek/SambaNova)
        ↓
AI requests tool calls (read_script, write_script, etc.)
        ↓
Server sends tool requests to plugin (HTTP GET /poll)
        ↓
Plugin executes in Studio, posts results back
        ↓
Server feeds results to AI, gets final response
        ↓
Plugin displays AI response in chat

No Claude Code CLI required. The server is standalone — just run it and chat.

Quick Start

Prerequisites

• Node.js v18+

• Roblox Studio

• One free API key (Groq, Gemini, Claude, Mistral, DeepSeek, SambaNova, or HuggingFace) — or Ollama for 100% local use

Setup

Windows (one-click):

setup.bat

Manual:

npm install
npm run build

Copy the plugin to Roblox Studio:

# Windows
copy plugin\ClaudeMCP.lua "%LOCALAPPDATA%\Roblox\Plugins\"

# macOS
cp plugin/ClaudeMCP.lua ~/Documents/Roblox/Plugins/

Run

Step 1 — Enable HTTP Requests in Roblox Studio:

In Roblox Studio, go to Game Settings > Security and turn on "Allow HTTP Requests". This is required for the plugin to talk to the server.

Step 2 — Start the server:

Double-click start.bat in the project folder. A terminal window will open and the server will start instantly.

Or manually in a terminal:

cd "MCP Claude"
node build/index.js

You should see:

  Roblox Studio AI Server
  ========================
  Server running on http://localhost:3636
  Waiting for Roblox Studio plugin to connect...

Keep the terminal window open — the server must be running while you use the plugin. Close it to stop the server.

Step 3 — Use it in Roblox Studio:

1. Open Roblox Studio — the plugin loads automatically with an "AI Assistant" panel

2. The status dot should turn green (connected to server)

3. Click the gear icon in the plugin panel

4. Enter your API key (or configure Ollama URL)

5. Select an AI provider from the dropdown

6. Start chatting!

Troubleshooting: If the status dot stays red:

• Make sure start.bat is running (the terminal window is open)

• Make sure "Allow HTTP Requests" is enabled in Game Settings > Security

• The server must be running BEFORE you open Studio

Environment Variables (Optional)

You can also set API keys via .env file:

cp .env.example .env
# Edit .env with your keys

What the AI Can Do

Example Prompts

• "Create a part in Workspace called MyPart"

• "Show me all scripts in ServerScriptService"

• "Make all parts in the model red and anchored"

• "Create a sprint script for the player"

• "Find all Part instances and make them transparent"

Plugin GUI

The dockable panel in Studio has:

• Provider dropdown — switch between Groq, Gemini, Ollama, HuggingFace, Claude, Mistral, DeepSeek, SambaNova

• Settings (gear icon) — configure API keys and Ollama URL

• Chat area — conversation with the AI, tool call notifications

• Status dot — green = connected, red = disconnected, yellow = thinking

Chat Commands

Architecture

Server (src/)

• index.ts — Entry point, starts Express server on port 3636

• bridge.ts — HTTP endpoints: /chat, /poll, /result, /providers, /config, /health

• tools.ts — 11 tool definitions (JSON Schema format)

• ai/provider.ts — AI provider interface

• ai/groq.ts — Groq API client

• ai/gemini.ts — Google Gemini API client

• ai/ollama.ts — Ollama local API client

• ai/huggingface.ts — HuggingFace Inference API client

• ai/claude.ts — Anthropic Claude API client

• ai/mistral.ts — Mistral AI API client

• ai/deepseek.ts — DeepSeek API client

• ai/sambanova.ts — SambaNova API client

• ai/chat.ts — Chat engine with tool execution loop

Plugin (plugin/)

• ClaudeMCP.lua — Full plugin with chat UI, provider selector, settings panel, and tool execution

Communication

Roblox Studio can't host HTTP servers, so the plugin polls:

1. User sends message in plugin chat

2. Plugin POSTs to /chat on the server

3. Server calls the AI provider

4. AI returns tool calls → server queues them

5. Plugin polls GET /poll, executes tools, POSTs results to /result

6. Server feeds results back to AI until done

7. Final response sent back to plugin

Development

# Watch mode
npm run dev

# Build
npm run build

# Start
npm start

Change port:

PORT=4000 npm start

License

MIT