Blender gets an MCP server and AI assistant add-on
An MIT-licensed Blender extension pairs a chat workspace with 21 scene tools, a built-in MCP server and optional manager-worker-reviewer agent chains.
A GitHub project called hitkeeen/blender-ai-mcp-addon now ships a Blender MCP server alongside an AI chat assistant that runs inside Blender itself. According to the project's GitHub repository, the add-on, named mcp-for-blender-pro, was created on 2026-11-12 and had 301 stars and 3 forks as of 2026-11-14. It is released under the MIT license, and the release bundle targets Windows x64 with Blender 4.2 or later.
A chat panel, a Studio window and 21 scene tools
The add-on adds a panel to Blender's 3D viewport sidebar. From there a user picks a provider, enters connection details and selects a model, then sends messages without leaving Blender. A separate Studio window shows the conversation next to scene objects, viewport snapshots, approval prompts and a task board.
The source lists 21 tools grouped by job. Scene inspection covers reading scene info, listing objects and getting object details. Creation covers cubes, spheres, cylinders, planes, lights and cameras. Editing covers moving, duplicating, deleting, parenting and setting materials. There are also tools for capturing a viewport screenshot, running Python code, looking up the Blender Python API, describing node types, exporting a scene, and searching or importing assets. Asset integrations named in the source are Poly Haven, Sketchfab and Poly Pizza, and the project says to follow the licenses of whatever assets are imported.
The README gives sample requests rather than demos, for example asking the assistant to inspect a scene and explain its object structure before changing anything, or to build a product-display setup with a pedestal, a camera and three lights, with a matte blue material on the pedestal.
Choosing models and running agent teams
Supported connections listed in the source are OpenAI, Anthropic Claude, Google Gemini, OpenRouter, OpenCode Zen, Ollama and custom APIs. Custom endpoints can speak the OpenAI Chat Completions, Anthropic Messages or Ollama protocols. The project advises choosing models that support tool calling, and notes that image-capable models can read viewport screenshots. Saved connections are kept separate, so work and personal keys can coexist.
Two modes are described. Single assistant is for direct conversational edits. A Team chain splits work between a manager that writes assignments and acceptance criteria, workers that perform scene operations, and reviewers that approve or return findings. Tasks run sequentially so edits stay ordered. A blocked task gets one correction attempt, and a second failed review stops the workflow.
Install, requirements and what is left open
Installation goes through Blender's preferences: open Get Extensions, choose Install from Disk, select the ZIP from Releases, enable the add-on and open the MCP Pro tab in the viewport sidebar. Blender must stay open while Studio or an external client is connected.
The built-in MCP server uses authenticated Streamable HTTP, and the project says a bundled OpenCode Portable client can be launched from the sidebar with the Blender connection already configured. Other compatible MCP clients can connect to the local endpoint described in the connection guide.
On safety, the source is explicit: Python tool validation helps catch disallowed operations but is not a security sandbox. Operation approvals, cancellation after the current step and named undo steps exist for built-in chat, while external MCP clients manage their own approval policies. Credentials are stored in user settings and chat history lasts only for the current session.
Open questions the source does not answer: whether macOS or Linux builds exist, what a team chain costs in tokens for a long task, and how the add-on behaves on heavy scenes. A fast star count is a signal of attention, not of reliability.
What this suggests for people building apps
The project is independent and states it is not affiliated with or endorsed by the Blender Foundation or the listed AI providers. What it illustrates is a pattern that is likely to spread: wrap a tool's real operations as discrete, inspectable functions, add approval and undo, then expose the same functions over MCP so different clients can drive them. If you are wiring AI features into your own app, that pairing of narrow tools and explicit approval steps is worth copying. Draftship's notes on adding AI to your app cover the app-side version of the same decision.
Primary sources
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.