Table of Contents Show
For those who don’t know the tool: 3DGROUND develops scripts and workflow utilities for Autodesk 3ds Max, while maxpkg handles installation and updates for Max tools. Max Ultra MCP adds a local Model Context Protocol bridge, allowing compatible AI clients to inspect and operate an already-open 3ds Max scene instead of merely suggesting scripts from the sidelines.
Max Ultra MCP connects MCP-compatible AI assistants directly to an open Autodesk 3ds Max session, giving them tools to inspect scene state, execute MaxScript, manipulate objects, capture the viewport and control supported renders. Crucially, the assistant can then inspect what actually happened before taking the next step, rather than firing a script over the wall and hoping for it to stick. If artificial intelligence is finally a match for biological stubbornness, 3ds Max Users may have found their sparring partner.
The tool is developed by Lukianenko Vasyl / 3DGROUND and distributed through MaxPkg. The current package supports Windows and 3ds Max 2022 through 2027. The developer documents Codex, ChatGPT Desktop and Claude Code as setup targets, with other clients possible if they support local MCP servers over STDIO.
MCP, meet Max
The important bit is the Model Context Protocol, not a new AI model. Max Ultra MCP exposes 3ds Max functions as tools that an AI client can call. Underneath, JSON-RPC over STDIO connects the AI client to a local MCP host. A bridge daemon then communicates with a MaxScript bootstrap running inside each connected 3ds Max process. Operations touching the scene, viewport, renderer or Max-owned UI are routed back through Max. This is considerably more useful than asking a chatbot to write MaxScript and then becoming its error-message courier.
A tool belt, not just a text box
Max Ultra MCP provides dedicated tools for common scene operations. It can create, select, rename, transform, clone and delete objects, build meshes and splines, add modifiers, work with layers, cameras, lights and basic animation, and inspect scene information including units, materials and renderer state.
When the dedicated tools are not enough, it can also execute arbitrary MaxScript, files, macros and 3ds Max Action commands. That flexibility comes with the obvious production footnote: unrestricted MaxScript can make unrestricted changes. The developer recommends backups before large operations, and the repository treats arbitrary script execution as a destructive operation that AI clients should place behind an approval step.
The viewport gets a vote
Max Ultra MCP can capture the active viewport and return the image to the connected AI client. It can temporarily maximize the viewport and hide overlays before capture, then restore the previous display state. That lets a vision-capable model inspect the result of its own work. An agent might build geometry, capture the viewport, compare the image with the requested result and make another adjustment.
The developer’s floor-plan example takes this further. A drawing supplied to the AI client is interpreted into structured dimensions, validated and then used to build geometry in Max. The source image itself is not simply handed to 3ds Max.
This split is sensible. Language-model interpretation remains fuzzy around the edges, while the actual scene mutation can still pass through explicit data and validation. The robot may improvise, but at least the walls can have measurements.
Local bridge, cloud manners
Communication between the MCP bridge and 3ds Max stays on the local machine. The developer says listeners bind to 127.0.0.1 and local requests use a random installation token. That does not automatically make the complete workflow offline. Cloud AI clients still require their usual network connection, and scene information, screenshots or tool results requested by the model can leave the workstation through that client.
The developer says Max Ultra MCP does not automatically upload the complete .max file. Even so, studios should review the data policies of whichever AI service sits at the other end of the conversation. “Local bridge” and “local inference” remain very different beasts.
Renderers remain renderers
The bridge can start, monitor, cancel and retrieve supported renders, as well as inspect renderer information. More ambitious renderer-specific automation is still developing. The repository lists dedicated Corona and V-Ray configuration adapters among planned work rather than finished semantic tools.
That distinction matters. Max Ultra MCP already exposes useful render control, but it should not yet be presented as a universal AI configuration layer for every renderer and every installed version. In other words, the AI can talk to rendering. Rendering has not agreed to listen, but …
Multiple Maxes, fewer accidents
Several running 3ds Max instances can connect to the bridge. When only one is present, selection can happen automatically. With several open sessions, the MCP client must choose which one it intends to operate. That is unglamorous but valuable pipeline plumbing. An automation system that confidently edits the wrong scene is still confident.
Interrupted mutation calls are also not automatically replayed because an operation may already have reached Max before communication was lost. That avoids turning a network hiccup into duplicate geometry, duplicated modifiers or a small archaeology project.
Installation without interpretive dance
Max Ultra MCP ships as an MZP package and can be installed through MaxPkg Manager, from the MaxPkg website or manually. A portable Node.js runtime is bundled, so normal installation does not require a separate Node.js setup, PATH editing or administrator rights.
The onboarding panel can configure supported AI clients through their command-line tools. If the expected command is unavailable, it shows the STDIO connection details for manual configuration rather than quietly rewriting unrelated settings. The current v1.2.5 package supports 3ds Max 2022 through 2027. Exact renderer and plug-in behaviour can still depend on the versions installed alongside Max.
Production-ready-ish
There is one status wrinkle. The MaxPkg catalog labels Max Ultra MCP “Production-Ready.” The developer has also described the project publicly as open for testing and still evolving.
The repository README is more cautious again. It says the v1 architecture and mock-tested MCP surface are implemented, while real 3ds Max and renderer-version acceptance testing is still required before a production release. Several planned workflows remain on the backlog.
The safest reading is therefore straightforward: the bridge is functional and already broad in scope, but studios should validate the precise host, renderer, plug-in and AI-client combination they intend to use rather than treating the catalog badge as a substitute for pipeline testing.
Free, with paperwork still invited
MaxPkg currently lists Max Ultra MCP as Free, and the developer has also described it as completely free. The source is publicly visible on GitHub, but a public repository alone does not establish an open-source licence. The material reviewed for this article does not provide enough basis to call Max Ultra MCP open source or infer specific commercial-use rights.
https://maxpkg.dev/catalog/max-ultra-mcp
https://github.com/maxpkg-dev/max-ultra-mcp
https://3dground.net/en/
| Product | Max Ultra MCP |
| Developer | Lukianenko Vasyl / 3DGROUND |
| Release status | Public testing according to the developer; MaxPkg also labels it Production-Ready |
| Version | 1.2.5 |
| Host compatibility | Autodesk 3ds Max 2022–2027 |
| Operating system | Windows |
| AI clients | Codex, ChatGPT Desktop, Claude Code; other MCP-compatible STDIO clients |
| Connection | Local loopback bridge; cloud AI clients generally require internet |
| Price | Free |
| Documentation | GitHub repository |
| Download | MaxPkg |
