Connect an AI tool to the Studio (MCP)
Let Claude Code or any MCP client drive the Studio you have open in your browser, with a token you make in Settings, scopes you choose and calls you allow.
On this page
Every button, menu and panel in the Studio runs a named command, and every command is a tool your own AI assistant can call. Through the Model Context Protocol (MCP), a tool such as Claude Code can look at your scene, add and change objects, edit files, take screenshots and play your game in the Studio you have open in your browser, as you would.
The connection goes through the NEO Platform: your tool talks to the platform with a token you make, and the platform passes each call to your open Studio. You choose what each token may do, you can see every call it makes, and you can revoke it at any time.
Turn on MCP access in the Studio#
- Open the Studio in your browser, signed in to your NEO account.
- Open the MCP Connections panel and choose Accept MCP clients. The panel shows that the Studio is connected and lists the clients that call it.
The Studio keeps accepting clients until you turn it off there or close the tab. You can have more than one Studio open; each is its own session.
Make a token#
In Settings → MCP access, choose Make a token:
- Who it's for: a name you'll recognise later, such as "Claude Code on my laptop".
- What it may do: its scopes (below).
readis always included. - Expires: never, or after a number of days.
The token is shown once. Copy it straight into your tool's settings and keep it like a password: anyone who has it can drive your Studio within its scopes. The platform keeps only a fingerprint of it, so it can't show it again; if you lose it, revoke it and make another.
Scopes that let a tool touch your project's files, delete, publish or start processes ask you to confirm it's you first, with your passkey or authenticator app, or your password if your account has neither.
Scopes#
| Scope | What it allows |
|---|---|
read | Seeing the Studio: the scene, the selection, project files, screenshots, the log |
edit | Changing the scene and the editor: adding, moving and editing objects, selecting, opening panels (all undoable) |
fs | Reading and writing your project's files |
build | Building and packaging the project |
destructive | Deleting or overwriting objects, assets and files |
publish | Publishing the game and inviting testers |
process | Starting and stopping processes the Studio runs |
secrets.write | Setting API keys and other secrets in the Studio |
A call that needs fs, build, secrets.write or edit and that the token lacks is refused, and the tool is told which scope it needs. A call that needs destructive, publish or process and that the token lacks isn't refused: it waits for you.
Calls that wait for you#
When a tool asks for something its token isn't allowed to do on its own, such as deleting an object or publishing, the Studio holds the call and shows it in the MCP Connections panel under Waiting for you, with the tool's name and what it asked for. Choose Allow to run it once, or Deny. The tool gets the result, or a message saying you declined.
Only you, in the Studio, can answer a waiting call. No token can approve one, not even one with every scope, and the platform has no way to approve one for you. A waiting call gives up after a few minutes.
Connect Claude Code#
Settings shows a ready-made command after you make a token. It looks like this:
claude mcp add --transport http neo-studio https://api.neoengine.io/mcp/studio \
--header "Authorization: Bearer neomcp_..."Then ask Claude to look at your Studio, for example "list the objects in my scene" or "add a cube above the player and take a screenshot".
Connect another MCP client#
Any client that speaks MCP's Streamable HTTP transport works. Point it at the endpoint Settings shows, with the token as a bearer token:
{
"mcpServers": {
"neo-studio": {
"type": "http",
"url": "https://api.neoengine.io/mcp/studio",
"headers": { "Authorization": "Bearer neomcp_..." }
}
}
}The server is named neo-studio. Its tools are the Studio's commands with dots turned into underscores: scene_list, editor_state, editor_screenshot and so on. A tool lists only what its token can call (plus the ones that wait for you).
Pick a Studio#
When you have more than one Studio open, calls go to the one you opened most recently. To choose:
studio_sessionslists your open Studios, with their ids (mss_...), projects and which one this connection uses.studio_use_sessionwith asessionid makes the connection drive that Studio.- Or connect to a Studio's own address,
https://api.neoengine.io/mcp/studio/mss_..., shown in Settings next to each open Studio.
A token only ever reaches your own Studios.
Limits#
To keep the Studio and the platform responsive, each token may make up to 20 calls a second, 600 a minute and 20,000 a day, and your account up to 1,200 a minute across its tokens. A request may be up to 8 MiB and an answer up to 12 MiB: ask for screenshots at a smaller size, and move big files in parts. A call that the Studio doesn't answer within a minute fails, unless it is waiting for you.
Too many wrong tokens from one network pause sign-ins from it for a few minutes.
See and revoke access#
Settings → MCP access lists your tokens with their scopes, when they were made and last used, and your Studios with the tokens that used them. Revoke stops a token at once: its next call is refused and any of its calls waiting in a Studio are dropped.
Every token you make is also in your account's security activity, and you get an email about it. If you didn't make it, choose This wasn't me in that email or in Settings → Security: every device is signed out and every MCP token is revoked. Signing out everywhere and changing your password revoke your MCP tokens too.