precygtrucqui1981/blender-ai-copilot-lab

What it is

blender-ai-copilot-lab is a Blender extension that turns a scene brief into a supervised experiment instead of a single chat thread. A manager agent splits the brief into assignments with acceptance criteria, worker agents carry them out as ordered scene operations, and reviewer agents approve each result or return specific findings. The run is visible on a live task board, with a named undo step for every mutation.

It is aimed at Blender users who want multi-step scene construction (props, cameras, lighting, materials) to stay structured, reviewable and reversible rather than implied by one long prompt.

Quick start

1. Download mcp-for-blender-pro.zip from the latest release.
2. In Blender, open Edit → Preferences → Get Extensions → menu → Install from Disk.
3. Select the ZIP, enable mcp-for-blender-pro, then open 3D Viewport → N → MCP Pro.
4. Choose a provider, add connection details and select a model.
5. Send a message in Blender or click Open Studio Window.

To use the team mode, switch the workflow mode from Single assistant to Team chain and configure exactly one manager, one or more workers and one or more reviewers. Each agent has its own name, instructions, provider and model; blank provider/model fields inherit the workspace connection.

Configuration and MCP endpoint

Supported providers are OpenAI, Anthropic Claude, Google Gemini, OpenRouter, OpenCode Zen, Ollama and a Custom API. Agents can mix providers, for example a local Ollama model for workers and a cloud model for review.

The extension can expose a local Streamable HTTP MCP endpoint at http://127.0.0.1:9876/mcp endpoint setting, enabled in preferences; a bearer access token is generated when it is enabled and should be stored in the client's secret store. The interface language can be set to English, German, French, Spanish or Chinese.

Constraints

• The extension is labeled for Windows x64 and Blender 4.2+ platform badges.
• Tasks run sequentially, not in parallel, so scene edits stay ordered and reviewable.
• Exactly one manager can be enabled; the chain refuses to run with two.
• A blocked task gets one correction attempt; a second failed review stops the chain after the current step.
• A single custom endpoint is shared by all custom-protocol agents.
• Released under the MIT license license.

HotRepo's take

The repository was created on 2026-11-12 and recorded 100 stars and 2 forks in the new-project snapshot covering roughly the first 15 hours after creation, making it an early-stage project with a single observation point. That number reflects the launch window rather than sustained adoption. The README describes a fairly detailed agent-workflow design, but GitHub detected no programming-language code in the repository, and the usable artifact is a ZIP installed through Blender; the task board, review findings and undo behavior described here have not been independently exercised in this review.