Octane is becoming something you can talk to, build with, and control.
I’ve been pushing the new MCP inside Standalone across procedural worlds, Vectron, shaders, and even custom simulation workflows.
The full Octane MCP guide is now live on OTOY. https://x.com/i/article/2118087331836022785
OCTANE + MCP: WHEN A RENDERER BECOMES A WORLD BUILDER
The new @OTOY Octane MCP guide is live, and things are getting interesting.
Over the last few days I’ve been pushing the new MCP integration inside Octane Standalone to see what happens when an agent can work directly with the renderer.
The answer is: quite a lot.
I built a fully procedural garden labyrinth using OSL, Vectron and code-based geometry. The scene represents more than 2 billion polygons while the procedural scene itself is only around 360 KB.
Then I pushed Vectron further with a brutalist ruin generator built entirely from code. No modeled geometry. No texture maps. Every surface is procedural and seedable.
Then I went outside Octane entirely, built a pyro solver from scratch with Claude, and created a graph inside Octane that could bring those frames back in as native, art-directable volumes.
These experiments all came from the same underlying idea:
What happens when Octane becomes something an agent can actually work inside?
Octane Standalone 2027.1 now includes a local MCP server that gives compatible agents access to Octane through its Lua API.
That opens up scene construction, node graphs, OSL shaders, Vectron geometry, materials, lights, cameras, rendering, custom tools and more.
But the interesting part isn’t simply automating clicks.
The agent can write something, execute it inside Octane, inspect what happened, respond to errors, revise it, render the result and continue iterating.
And when it’s finished, you’re still working with an Octane scene.
You can grab the controls yourself. Change the parameters. Art direct the result. Package it as ORBX. Bring it into an Octane-supported DCC. Keep working.
That’s the part I think is easy to underestimate.
Natural language becomes another way into the renderer, and suddenly ideas that might once have required searching for a plugin, writing a tool from scratch or abandoning an experiment entirely become much more approachable.
I put together a full guide on the OTOY Help Center covering how to connect Claude Code, Codex, Cursor, VS Code, Gemini CLI, Claude Desktop, ChatGPT Desktop and other compatible agents directly to Octane Standalone.
It also gets into what the connection can actually do once you’re inside.
Connecting AI Agents to Octane Standalone (MCP)
https://help.otoy.com/hc/en-us/articles/58194506495131-Connecting-AI-Agents-to-Octane-Standalone-MCP
And if you want to see a simpler look-development example in action, this MCP design study shows direct shader manipulation and iteration inside Octane:
I’ve barely scratched the surface.
More experiments coming.
X Article caption
Octane is becoming something you can talk to, build with, and extend.
I’ve been pushing the new MCP inside Standalone across procedural worlds, Vectron, shaders and even custom simulation workflows.
My full Octane MCP guide is now live on OTOY.