Blender gets AI assistant with multi-agent workflows

A transparent holographic 3D workspace suspended in a dark studio, with fragmented geometric shapes orbiting a central mesh, connected by glowing threads of light that pulse in sequence—symbolising AI

A new open-source addon for Blender, mcp-for-blender-pro, introduces an AI assistant capable of inspecting scenes, generating objects, assigning materials and managing complex workflows—all from within the 3D workspace. Developed independently of the Blender Foundation, it integrates with popular AI providers including OpenAI, Anthropic Claude, Google Gemini, OpenRouter, OpenCode Zen, Ollama and custom API endpoints. This allows users to automate repetitive tasks or prototype scenes using natural language.

The addon includes a visual chat interface called Studio, accessible via a sidebar panel in the 3D Viewport (opened with N → MCP Pro), where users can track AI actions, review viewport snapshots and approve changes. It supports multilingual menus (English, German, French, Spanish, Chinese) and runs locally. Chat history is limited to the current session, and connection credentials are stored in user preferences. A built-in Model Context Protocol (MCP) server enables authenticated connections from external clients such as OpenCode.

How the AI assistant works

Users begin by connecting a model—via a cloud provider, local Ollama instance or custom API endpoint. After configuration, they can issue commands like ‘inspect this scene’ or ‘create a product display with a blue pedestal and three lights’. The assistant uses 21 built-in scene tools to execute tasks, including:

  • Scene inspection: get_scene_info, list_objects, get_object_info
  • Object creation: add_cube, add_sphere, add_cylinder, add_plane, add_light, add_camera
  • Editing: move_object, duplicate_object, delete_object, parent_objects, set_material
  • Visual feedback and scripting: viewport_screenshot, execute_code
  • Blender API and export: bpy_api_lookup, describe_node_type, export_scene
  • Asset workflows: search_assets, import_asset

The system supports both direct interaction via single-model chats and structured multi-agent workflows. In team mode:

  • A Manager model breaks down requests into subtasks with acceptance criteria.
  • Worker models perform the assigned scene operations.
  • Reviewer models inspect results and either approve or return findings.

Tasks run sequentially to maintain order in scene edits. If a task fails, it gets one correction attempt; if the review fails a second time, the workflow stops. The task board in Studio displays assignments, attempts and review feedback in real time.

Flexibility in AI model choice

One of the addon’s core strengths is flexibility in AI model selection. Users can switch between providers without leaving Blender, selecting models with tool-calling and image-input capabilities directly from a loaded catalog. Image-capable models can interpret viewport screenshots, enabling contextual understanding of the scene.

Local models via Ollama reduce reliance on external APIs, supporting data sovereignty and offline operation. Custom endpoints support OpenAI Chat Completions, Anthropic Messages and Ollama protocols. OpenCode-only free models are accessible through the OpenCode client.

OpenCode Portable can be launched directly from the addon’s sidebar, with the Blender connection preconfigured. OpenCode manages its own authentication and model selection, making it suitable for users concerned about cost, data control or vendor lock-in—particularly in regions like the UK, UAE, US and Kazakhstan, where cloud access, pricing or compliance requirements may vary.

Implications for 3D workflows

For artists and technical directors, the addon accelerates iteration, especially in product visualisation, scene layout and asset prototyping. Developers can integrate it into larger pipelines using MCP-compatible clients. However, Python tool validation is not a security sandbox: while it helps catch disallowed operations, users must remain cautious when executing untrusted code via the execute_code tool.

Asset integrations with Poly Haven, Sketchfab and Poly Pizza require users to configure API keys where needed and comply with each platform’s licensing terms. The addon streamlines access but does not enforce license compliance automatically. Users in regulated or commercial environments should verify usage rights, particularly when deploying outputs at scale.

What it means

This addon lowers the barrier to AI-assisted 3D creation without tying users to a single provider. Its local execution, multi-agent logic and support for offline models make it a compelling option for studios prioritising data control and operational resilience. For founders and IT teams, it represents a modular step toward automation—without surrendering autonomy to proprietary ecosystems.

The MIT-licensed tool is available for Windows x64 with Blender 4.2 or later. Installation is via ZIP import in Blender’s extension manager (Edit → Preferences → Get Extensions → Install from Disk). As an independent project, it receives no funding or endorsement from the Blender Foundation or AI vendors.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *