Roblox Executor MCP Server

An MCP server that allows Agents to interact with a running Roblox game client — execute code, inspect scripts, spy on remotes, and more.

Dashboard

Roblox Executor MCP includes a local web dashboard at:

http://localhost:16384/

Use it to see connected Roblox clients, inspect scripts, run tools, view server logs, configure semantic search, and index games for semantic script search.

Features

• Code Execution — Run Lua code and fetch data from the game client.

• Script Inspection — Decompile scripts and search across all sources.

• Instance Search — CSS-like selectors and hierarchy trees.

• Remote Spy — Intercept, log, block, and ignore Remotes/Bindables via Cobalt.

• GUI Interaction — Click buttons and type into text boxes.

• Screenshot — Capture Roblox window screenshots (Windows only).

• Multi-Client — Connect multiple Roblox clients at once.

• Primary / Secondary — Multiple MCP instances auto-coordinate with automatic promotion. Supports remote relaying via --baseurl. See Advanced.

Tutorial

Prerequisites

• Node.js ≥ 18

• Bun ≥ 1.3 for the interactive OpenTUI harness installer

• A Roblox executor that supports loadstring, request, and (preferably) WebSocket

Quick Start

1. Clone the server

git clone https://github.com/dissering/roblox-executor-mcp.git
cd roblox-executor-mcp

2. Run the harness installer

The installer builds the server, lets you choose AI clients, writes supported MCP configs, and prints the Roblox loader script.

npm run install:harnesses

The normal server build consumes the committed connector.luau artifact, so harness installs do not require Darklua. Connector developers can install the pinned tool with rokit install, edit connector-src/, and regenerate the artifact with npm run build:connector.

The picker is built with OpenTUI and runs through Bun. npm run install:harnesses installs Bun first if it is not already available. It shows detected local clients by default; if none are detected, it warns you to install a harness first. Press s in the picker or pass --show-all-harnesses to reveal every supported config target. If your terminal has trouble with the interactive picker, use the plain numbered prompt:

npm run install:harnesses -- --plain

The installer can also place the Roblox loader into a detected executor autoexec folder, such as MacSploit on macOS or supported Windows executor autoexec folders. Use the prompt, or run:

npm run getscript -- --autoexec

It can also help with:

• cross-machine setup on the same LAN

• copying the Roblox loader to your clipboard

• optional Ollama embeddinggemma setup for semantic indexing

• pulling latest repo changes before install/build

To update an existing install later, run:

npm run update

The update command can stop currently running MCP server processes, optionally pull latest changes, and always rebuilds the server.

Manual setup

If you prefer to configure a client yourself, use the setup guide for your client:

3. Connect from Roblox

The installer prints this for you. Put it in your executor or Auto Execute:

while not getgenv().MCP_Loaded do
    local bridgeUrl = getgenv().BridgeURL or "localhost:16384"
    pcall(function() loadstring(game:HttpGet("http://" .. bridgeUrl .. "/script.luau"))() end)

    task.wait(0.15)
end

Optional settings (set before the loadstring):

getgenv().BridgeURL = "10.0.0.4:16384"                  -- default: localhost:16384
getgenv().DisableWebSocket = true                        -- force HTTP polling
getgenv().DisableInitialScriptDecompMapping = true       -- skip initial decompilation
getgenv().MCP_FailedScriptResyncInterval = 30            -- retry failed script syncs periodically
getgenv().MCP_FailedScriptResyncBatchSize = 8            -- bound each periodic retry batch

After the MCP server starts and Roblox connects, open the dashboard:

http://localhost:16384/

Community

Have a suggestion or need help? Join the Discord server.

Security

This server allows arbitrary code execution. Only use with AI clients you trust. Port 16384 has no authentication — never expose it to the internet. For cross-machine setups, use a local network, VPN, or SSH tunnel. See Advanced for details.

License

MIT