Roblox

You can control Roblox Studio from AI coding tools using an MCP (Model Context Protocol) server. This setup lets you explore the workspace, manipulate instances, manage scripts, integrate toolbox assets, run playtests, and render viewport previews, enabling automation and rapid iteration from your AI workflows.

How to use

You will run the MCP server locally and connect an MCP client (for example, an OpenCode or similar AI tool) to issue commands. Start the server, then configure your client to point to the local bridge. Use the provided tools to inspect and mutate the Roblox Studio scene, manage scripts, search and insert toolbox assets, and control playtests and camera views. Commands are issued through a consistent action interface, with responses returned from Studio after execution.

How to install

Prerequisites: you need Node.js installed on your machine. You also need Roblox Studio installed if you want to connect to a live Studio session.

# 1) Install dependencies for the MCP server
npm install

# 2) Build the MCP server bundle
npm run build

Configuration and usage notes

The MCP server exposes an HTTP bridge on localhost port 28821 and provides a local runtime entry for the client to start a server process. The client should be configured to communicate with 127.0.0.1:28821 for command polling and result reporting, and the server process should be started with the appropriate runtime command and environment.

Run the local server process using the command shown in the setup. The bridge port can be customized via environment variables if needed, but the default port is 28821.

Additional sections

Instance paths are addressed using dot-separated names from the global game object, such as game.Workspace, game.Workspace.SpawnLocation, or game.ServerScriptService.ModuleScript. These paths help you target specific instances for reading properties, updating values, or performing mutations.

Environment variable you may set is ROBLOX_MCP_PORT to designate the HTTP bridge port. The default is 28821.

Known limitations include that screenshots are software-rendered and not pixel-perfect, playtest behavior may differ from local Play Solo, toolbox searches may be rate-limited by the Roblox API, and script execution runs in a temporary ModuleScript context which may not expose all studio APIs.

Tools and actions overview

The MCP server exposes a set of tool groups for reading data, mutating instances, managing scripts, controlling the scene camera and screenshots, manipulating toolbox assets, and running playtests. Each tool requires an action parameter to select the specific operation.

Troubleshooting

If the bridge cannot connect, verify that Roblox Studio is running and the MCP Bridge button in Studio indicates a connection. Ensure the MCP port (default 28821) is not blocked by local firewalls and that the client is configured to reach 127.0.0.1:28821.

Security considerations

Only run the MCP server in trusted environments where the controlling AI has access to your Roblox Studio projects. Do not expose the bridge port to untrusted networks. If you enable remote clients, ensure proper authentication and access controls are in place.

Available tools

roblox_get

Read-only queries to retrieve information about the Roblox Studio workspace, instance tree, specific instances, properties, outputs, and other read operations.

roblox_manage

Mutate the Roblox Studio instance by creating, updating, cloning, reparenting, deleting, and selecting instances, plus undo/redo actions.

roblox_script

Create, read, update, and execute scripts within Studio contexts to manage code assets and run Lua snippets.

roblox_scene

Control the camera and capture screenshots of the current viewport, including auto-framing on a target instance.

roblox_toolbox

Search and insert assets from the Creator Store, and strip scripts from inserted models when needed.

roblox_playtest

Manage playtest sessions, control the play state, move cameras during tests, and trigger interactions like ClickDetectors and ProximityPrompts.