Build and Automate Blender Scenes with mcp-for-blender-pro

What is mcp-for-blender-pro?

mcp-for-blender-pro is an AI assistant and Model Context Protocol server for Blender. It lets an AI model inspect your scene, create and edit objects, assign materials, search for assets, capture viewport images, consult Blender APIs, and export scenes.

The extension is designed for artists seeking faster iteration, technical artists automating repetitive work, and developers connecting MCP-compatible tools to Blender. You can work through Blender’s sidebar, open the companion Studio window, or connect a compatible external MCP client.

The Studio workspace combines conversation history, scene objects, viewport snapshots, operation approvals, and a task board. Blender remains open and usable while the assistant works.

Key features

• Blender-aware assistance: The assistant can inspect scenes and individual objects before making changes.
• Visual Studio workspace: Review conversations, scene data, task progress, approvals, and viewport previews in a companion window.
• Multiple AI providers: Use OpenAI, Anthropic Claude, Google Gemini, OpenRouter, OpenCode Zen, Ollama, or compatible custom APIs.
• Saved connections: Maintain separate connection profiles, including multiple API keys for the same provider.
• Single and multi-agent modes: Complete quick edits with one assistant or divide larger jobs among managers, workers, and reviewers.
• Built-in MCP server: Connect OpenCode and other clients supporting authenticated Streamable HTTP MCP.
• Scene and asset tools: Use 21 tools for inspection, creation, editing, screenshots, scripting, API lookup, asset workflows, and export.
• Multilingual interface: The main menus support English, German, French, Spanish, and Chinese.

Requirements

The release bundle targets Windows x64 and requires Blender 4.2 or later. Download the mcp-for-blender-pro.zip release bundle rather than extracting it before installation.

You also need access to a supported cloud provider, a local Ollama installation, a compatible custom API, or OpenCode. Select a model that supports tool calling. If you want the assistant to interpret viewport screenshots, choose an image-capable model.

Install the Blender extension

1. Download mcp-for-blender-pro.zip from the project’s latest release.

2. Open Blender and select Edit → Preferences → Get Extensions.

3. Open the menu and choose Install from Disk.

4. Select the downloaded ZIP file.

5. Enable mcp-for-blender-pro after installation.

6. Return to the 3D Viewport, press N, and open the MCP Pro tab.

Keep Blender running whenever you use the Studio interface or an external MCP client because requests are sent to the local Blender connection.

Configure an AI connection

Open the MCP Pro interface and select a provider. Add the required connection details, load the provider’s available model catalog, and choose a model. Saved connections can separate personal and work credentials or preserve multiple accounts for one provider.

Supported connection options include:

• OpenAI
• Anthropic Claude
• Google Gemini
• OpenRouter
• OpenCode Zen
• Ollama
• Compatible custom APIs

Custom endpoints can use the OpenAI Chat Completions, Anthropic Messages, or Ollama protocol. OpenCode-only free models are accessed through the OpenCode client, which handles its own provider sign-in and model selection.

Use the assistant for your first scene task

Begin with an inspection request. Asking the model to explain the current scene before editing gives it useful context and lets you verify that it understands the object structure.

Inspect this scene and explain its object structure before making any changes.

Once the inspection is correct, request a concrete scene operation:

Create a simple product-display scene with a pedestal, a camera and three lights. Give the pedestal a matte blue material.

Follow the actions in the conversation and inspect the result in Blender or the Studio viewport preview. The assistant may use creation tools such as add_cube, add_light, and add_camera, followed by editing tools such as move_object and set_material.

Edit existing objects

For predictable edits, select the intended object in Blender and describe the arrangement precisely. For example:

Duplicate the selected object into a neatly spaced row of five and show the viewport.

This type of task can combine duplicate_object, move_object, and viewport_screenshot. Other editing tools can delete objects, create parent relationships, or change materials.

Search for external assets

The extension integrates with Poly Haven, Sketchfab, and Poly Pizza. Configure service keys where required, and always follow the license attached to each asset.

Find an appropriate Poly Haven asset for this scene and ask before importing it.

This wording separates discovery from import and explicitly requests approval. The relevant tools are search_assets and import_asset.

Choose between a single assistant and a team

Single assistant mode is appropriate for direct questions and quick scene changes. For a larger workflow, configure a Team chain with separate responsibilities, instructions, and models.

• Manager: Divides the request into assignments and defines acceptance criteria.
• Workers: Perform the assigned scene operations.
• Reviewers: Inspect the results and either approve them or return specific findings.

Team tasks run sequentially so edits remain ordered. If a task is blocked, it receives one correction attempt. A second failed review stops the workflow. Use the task board to monitor assignments, attempts, and review findings.

Writing a useful team request

Give the team a clear goal, constraints, and an expected final state. A structured request might look like this:

Create a product-display scene for the selected object.
Use a simple pedestal and three-point lighting.
Keep the existing object geometry unchanged.
Review the composition and material assignment before completion.

The manager can turn these requirements into ordered assignments, while reviewers use the constraints as acceptance criteria.

Understand the available Blender tools

The built-in tool set covers the following operations:

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

You do not normally need to name tools in your prompt. Describe the intended result clearly and let the assistant select the appropriate operations. Naming a specific tool can still be useful when you want to constrain a technical workflow.

Connect OpenCode or another MCP client

Use OpenCode Portable from the Blender sidebar to start OpenCode with the Blender connection configured automatically. OpenCode manages its own authentication and model selection.

Other compatible clients can connect to the authenticated local Streamable HTTP MCP endpoint. Follow the project’s connections and agents guide for endpoint configuration and authentication details. Keep Blender open while the client is connected.

Advanced workflow tips

Inspect before modifying

Start complex tasks by requesting get_scene_info, an object list, or details about important objects. This reduces ambiguity and gives you an opportunity to correct assumptions before mutations begin.

Use visual verification

Ask for a viewport screenshot after composition, lighting, or placement changes. An image-capable model can interpret the screenshot and use it during review.

Separate discovery from irreversible actions

For asset imports, deletion, scripting, and export, divide the workflow into inspection and execution stages. Ask the assistant to report what it found and request approval before proceeding.

Use focused acceptance criteria

In a Team chain, state measurable requirements such as object count, spacing, material assignment, camera presence, or export readiness. Specific criteria give reviewers a clearer basis for approval.

Use API lookup before custom scripting

For operations not covered by a dedicated scene tool, the assistant can consult Blender information through bpy_api_lookup and describe_node_type before using execute_code. Review generated operations carefully.

Safety, approvals, and session behavior

Built-in chat supports operation approvals, cancellation after the current step, and named undo steps for scene mutations. Python validation can detect some disallowed operations, but it is not a security sandbox. External MCP clients apply their own approval policies.

Connection credentials are stored in user settings, while chat history lasts only for the current session. Studio runs locally and authenticates requests sent to Blender.

Conclusion

mcp-for-blender-pro brings AI conversation, Blender scene tools, visual feedback, model selection, and agent workflows into one workspace. Start with a single assistant and inspection-first prompts, then adopt Team chains, asset integrations, viewport review, and external MCP clients as your workflows become more complex.

The project is released under the MIT license. It is independent and is not affiliated with or endorsed by the Blender Foundation or the listed AI providers.