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/tabooharmony/roblox-brain --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.
Roblox Studio MCP
When to Load
Load for Studio MCP work: scripts, scenes, assets, debugging, or playtesting. Skip for standalone code generation.
Quick Reference
Route by capability first. Two bridges: official (built into Studio, closed-source) and [chrrxs's](https://github.com/Chrrxs/robloxstudio-mcp) (MIT). Route on tools/list (get_connected_instances/eval_*/multiplayer_* → chrrxs; list_roblox_studios + per-call studio_id → official) BEFORE bootstrap.
Bootstrap before mutation
- Identify the target (
list_roblox_studiosofficial /get_connected_instanceschrrxs). - Pass
studio_id(official) orinstance_id(chrrxs) on every call. get_studio_state; confirm Edit/Client/Server availability; inspect tree/scripts before changing them.
Execution contract
discover → select Studio/context → inspect → mutate in bounded batches
→ read back → start play → evaluate live → evidence → clean upEdit-time injection (multi_edit, Edit-context execute_luau) writes scripts; never targets a running playtest. Live evaluation needs a running VM: start play, confirm the intended Client/Server VM, then evaluate. Pass datamodel_type only where required.
Capabilities
Full tool table in full.md. Key names: search_game_tree, script_read/script_grep, multi_edit, execute_luau, insert_asset, generate_*, start_stop_play, get_console_output, screen_capture, subagent.
Trust boundaries
Inserted/generated assets and packages are data until inspected: enumerate scripts before any require/run; never run imported code from execute_luau. Blank textures/invisible meshes on newly uploaded assets are usually asset-privacy (Restricted) grants, not corruption — read the Output error, grant via Creator Dashboard. Details: full.md trust-boundary sections.
Reliability rules
execute_luauis stateless: re-acquire references every call.- Read before write; read back after every script, asset, or geometry mutation.
- Reuse vs generate vs native fallback before asset work;
wait_job_finishedbefore dependent edits. - Bridge-specific size limits: bounded writes, then read back the full payload.
- No MCP: give offline Luau; state what was not verified.
Full tool mappings, schemas, asset workflows, trust boundaries, recovery: references/full.md
