How to use
- Copy the installation command and review the source package before installing it in your own agent environment.
- Check the source document for required tools, credentials and supported hosts.
- Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/el4cteo/roblox-skills --skill roblox-studio-mcpBefore you start
- Installation does not include a model subscription, API allowance or permission to access your files.
- Repository stars and installation counts describe activity, not an independent quality or security test.
Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.
Working in Roblox Studio through MCP
Roblox Studio ships a built-in MCP server (stdio). With it, an agent can explore the open place, edit scripts, run Luau in edit/server/client contexts, playtest, read the Output, capture the viewport, and simulate player input. It is the fastest way to verify Roblox changes instead of guessing.
Setup
- Latest Studio → open Assistant → … → Manage MCP Servers → enable Enable Studio as MCP server.
- Use Quick connect (Claude Code, Claude Desktop, Codex CLI, Cursor, Gemini CLI, VS Code, Antigravity)
or configure the client manually:
| OS | Command |
|---|---|
| macOS | /Applications/RobloxStudio.app/Contents/MacOS/StudioMCP |
| Windows | cmd.exe /c %LOCALAPPDATA%\Roblox\mcp.bat |
# Claude Code
claude mcp add Roblox_Studio -- /Applications/RobloxStudio.app/Contents/MacOS/StudioMCP
claude mcp add Roblox_Studio -- cmd.exe /c "%LOCALAPPDATA%\Roblox\mcp.bat" # Windows{ "mcpServers": { "Roblox_Studio": { "command": "/Applications/RobloxStudio.app/Contents/MacOS/StudioMCP" } } }Client-specific config (Codex config.toml, OpenCode opencode.json, Cursor/VS Code mcp.json): references/client-setup.md. A green indicator in Manage MCP Servers confirms the connection. Only connect clients you trust: they can read and modify your places.
Tools
| Group | Tools |
|---|---|
| Session | list_roblox_studios (every call takes a studio_id), get_studio_state (play state, available DataModels) |
| Explore | search_game_tree (filter by path/class/keyword, depth), inspect_instance (properties, attributes, children), script_search (fuzzy by name), script_grep (text across scripts, ≤ 50 hits), script_read (dot path like game.ServerScriptService.Main, line ranges) |
| Edit | multi_edit (several edits to one script, creates it if missing; datamodel_type = Edit) |
| Run | execute_luau (datamodel_type = Edit, Server, or Client; returns the result or error) |
| Playtest | start_stop_play, get_console_output, screen_capture (optional camera position/look-at), character_navigation, user_keyboard_input, user_mouse_input |
| Assets | search_asset (Creator Store / inventory), insert_asset, upload_image, store_image, generate_mesh, generate_material, generate_procedural_model + wait_job_finished |
| Delegation | subagent (explore or playtest types for multi-step investigations) |
| Knowledge | http_get (Roblox docs, API reference, Cloud docs), skill (Roblox-authored skills: debugging, device simulation, docs search, profiling, scene analysis, unit tests, rbx-open-cloud-usage, rbx-permissions, rbx-import-asset) |
| Open Cloud | open_cloud_invoke (any Open Cloud API, no API key; limited to the game's MCP permissions) |
MCP Permissions (Assistant → … → Settings → MCP Permissions, or the MCP status in Studio's bottom-right) are per user and per game: Open Cloud scopes, upload folders, data/memory store access from execute_luau, and the HTTP domain allow list. Only the user can turn them on; ask before anything touching money, players, secrets, or publishing, and never work around one that is off. Data store calls also need Enable Studio Access to API Services. Unsaved places can't get an HTTP allow list yet: save to Roblox first.
The working loop
- Orient:
list_roblox_studios→ pickstudio_id;get_studio_state. Map the project with
search_game_tree (depth-limited: services first), then script_grep/script_search for the feature.
- Read before writing:
script_readevery script you'll touch and its requirers. Check for a
Rojo/Script Sync setup (see below) before editing in Studio.
- Edit with
multi_edit(exact, minimal edits; keep existing style). Create instances/attributes/
tags with execute_luau in Edit context — changes made this way are real edits to the place.
- Verify statically: re-read the script; for pure modules,
execute_luaua quick check
(require the module in Edit and call functions).
- Playtest:
start_stop_play(start) → wait →get_console_outputfor errors → drive the scenario
with character_navigation / input tools → execute_luau in Server/Client to assert state → screen_capture for visual checks → start_stop_play (stop). Always stop play when done.
- Report what changed and what was verified (and what wasn't).
-- Example execute_luau (Server context during play): assert game state instead of eyeballing.
local Players = game:GetService("Players")
local results = {}
for _, player in Players:GetPlayers() do
local leaderstats = player:FindFirstChild("leaderstats")
local coins = leaderstats and leaderstats:FindFirstChild("Coins")
table.insert(results, `{player.Name}: {if coins and coins:IsA("IntValue") then coins.Value else "missing"}`)
end
return table.concat(results, "\n")Rules of thumb
execute_luauinEditmutates the place: prefer read-only queries while exploring, and batch
intentional changes into one clear script. Use ChangeHistoryService recording (TryBeginRecording/ FinishRecording) so the user can undo your edit as one step.
- Changes made during play (Server/Client contexts) are discarded when play stops — edit in
Edit. - Rojo projects: the file system is the source of truth — edit files on disk, let Rojo sync, and use
MCP to playtest/inspect. Editing synced scripts via multi_edit gets overwritten. Script Sync projects sync both ways, so either side works, but don't edit the same script in both at once.
- Keep
search_game_treequeries narrow (path + depth); large places produce huge outputs. - Multiple Studio windows: always pass the right
studio_id; tell the user which place you touched. - Inserted assets from the Creator Store can contain scripts: inspect them (
search_game_tree+
script_read) before keeping them; remove unexpected scripts (backdoors are common in free models).
- Don't claim success without evidence: cite console output, assertion results, or screenshots.
Roblox-authored skills and conversions
Studio exposes Roblox-authored skills through the skill tool (e.g. rbx-debug, rbx-perf-profiling, rbx-scene-analysis, rbx-unit-test, rbx-docs-search, rbx-device-simulator-lua). Roblox also publishes an official streaming conversion skill (/rbx-convert-to-streaming, downloadable from the "Techniques and conversion" streaming docs) that runs through this MCP server; back up the place first.
Related skills
roblox-testing (test runners), roblox-performance (profiling), roblox-tooling (Rojo/Script Sync), roblox-architecture (where things belong), roblox-code-review.