Codex + Blender MCP: A Practical Guide to Controlling Blender with AI

▶

Working in Blender with AI becomes much more interesting when the assistant does more than simply write Python code. It can also inspect the scene, check what it has created, capture viewport screenshots, and iterate on the model.

This is the goal of Blender Codex MCP: connecting Codex to an open Blender session using the Model Context Protocol, with a local add-on running inside Blender and an MCP server launched by Codex.

In practice:

1. Blender runs a local add-on.
2. The add-on opens a bridge on localhost:9876.
3. Codex starts an MCP server.
4. The MCP server forwards commands to Blender.
5. Codex can inspect, modify, capture, and prepare the scene for export.

The project is available on GitHub:

https://github.com/webita/blender-codex-mcp

If you would like to implement similar workflows in your company, Webita provides AI consulting, automation, and operational integration services:

Webita AI

Webita services: digitalization, AI, and automation

Free introductory AI course

Why Use MCP with Blender?

An AI model can already write Blender Python scripts. The problem is that, without feedback, it is working almost blindly.

With an MCP bridge, the workflow becomes far more effective:

1. Codex reads the scene.
2. Codex modifies an object or creates new geometry.
3. Codex captures a screenshot of the viewport.
4. The user visually checks the result.
5. Codex adjusts the camera, materials, lighting, or export settings.

This approach is particularly suitable for:

1. Technical mockups
2. Product scenes
3. Models for web configurators
4. Preparing .glb files
5. Repetitive Blender automations
6. Workflows requiring progressive visual review

The same principle also applies beyond Blender. Many companies can reduce repetitive work by connecting AI and automation to tools they already use every day, such as CRMs, websites, ticketing systems, spreadsheets, e-commerce platforms, and internal applications.

This is the type of solution Webita develops through its Webita AI service.

What Is MCP?

MCP, or Model Context Protocol, is an open standard for connecting AI applications to external systems, tools, data sources, and workflows.

The official documentation describes it as a standard for connecting AI applications to external tools and systems in a consistent way:

Model Context Protocol documentation

In this case, the external system is Blender.

What Does Blender Codex MCP Do?

Blender Codex MCP provides Codex with tools such as:

1. get_scene_info: reads the objects, materials, and current state of the scene
2. get_object_info: inspects an individual object
3. get_viewport_screenshot: captures a screenshot of the viewport
4. execute_blender_code: runs Python code inside Blender
5. blender_health_check: checks whether Blender is reachable
6. sync_camera_to_viewport: copies the current viewport view to the active camera
7. export_glb: exports the scene or selected objects as a .glb file

It does not replace Blender. It makes Blender controllable through Codex.

Requirements

Before you begin, you will need:

1. Blender installed
2. Codex desktop or a Codex runtime compatible with MCP
3. uv installed
4. Python 3.10 or later
5. The blender-codex-mcp repository

uv is a fast Python package and project manager. In this project, it is used to start the MCP server:

uv documentation

Installation

Clone the repository:

Then open Blender and install the add-on:

1. Open Blender.
2. Go to Edit > Preferences > Add-ons.
3. Click Install…
4. Select addon.py from the repository.
5. Enable Interface: Blender Codex MCP.
6. Press N inside the viewport.
7. Open the BlenderCodexMCP tab.
8. Click Connect to MCP server.

By default, Blender listens on:

Configuring Codex

Add the following configuration to ~/.codex/config.toml:

Replace /absolute/path/to/blender-codex-mcp with the actual path to the repository.

Restart Codex after updating the configuration.

First Test

With Blender open and the add-on connected, you can ask Codex:

Inspect the open Blender scene and take a viewport screenshot.

If everything has been configured correctly, Codex should be able to read the scene and use the MCP tools exposed by the server.

Recommended Workflow

The best way to work is not to ask Codex to rebuild the entire scene in a single step.

A better process is:

1. Inspect the scene.
2. Make a small, verifiable change.
3. Capture a viewport screenshot.
4. Adjust the camera, scale, materials, or lighting.
5. Perform the final export.

Examples of useful prompts:

Inspect the scene and tell me which objects are present.

Create a clean product-view camera and add soft studio lighting.

Center the selected model and place it on the ground plane.

Export the selected model as a GLB file for web embedding.

GLB Export and WordPress

The .glb format is one of the most practical options for publishing a 3D model on the web. It is compact, lightweight, and fully compatible with modern WebGL viewers.

A simple and effective workflow consists of a few essential steps.

Start by preparing the model in Blender. Optimize the scene by removing unnecessary elements and keeping the geometry clean. The model should then be correctly centered relative to the scene origin to ensure consistent positioning when displayed online.

Next, apply lightweight, optimized materials. Avoid excessively large textures that could slow down loading times on the web. It is also useful to configure an appropriate camera and lighting setup to generate an effective preview and present the model at its best.

Once optimization is complete, export the file in .glb format so that it is ready for use on the web.

Finally, upload the model to WordPress and display it through a WebGL viewer. This allows users to interact with the 3D object directly in their browser, without installing additional plugins or software.

A convenient solution for WordPress and Elementor is the <model-viewer> component:

model-viewer documentation

Example HTML:

The repository also includes a dedicated guide:

WordPress embed guide

If your goal is to integrate 3D models, configurators, automations, or dynamic content into a business website, you can also explore Webita’s website and AI services:

Webita AI

Webita: web agency, SEO, development, and automation

Codex Plugins: What Does That Really Mean?

A natural question is:

Can it be installed as a Codex plugin with one click?

At present, there is no official public marketplace for third-party Codex plugins. The repository also includes a local Codex plugin scaffold, which may be useful for future or experimental workflows, but it should not currently be presented as the primary installation method.

In short:

1. MCP server: the actual engine
2. Blender add-on: the bridge running inside Blender
3. Plugin scaffold: a future or local foundation for plugin packaging

Security

The execute_blender_code tool can execute Python code inside Blender.

This is powerful, but it requires caution:

1. Save your .blend file before making significant changes.
2. Work in small steps.
3. Check the results frequently using screenshots.
4. Only use repositories and scripts that you trust.

References

1. Blender Codex MCP repository
2. Webita AI
3. Webita digital services
4. Free introductory AI course
5. Model Context Protocol
6. Blender Python API
7. uv documentation
8. model-viewer

Blender Codex MCP is more than a simple experiment: it is a practical way to make Blender more accessible to AI-assisted workflows.

For professionals working with technical models, configurators, product scenes, or 3D content for websites, it introduces a practical and concrete workflow.

If you would like to evaluate an AI automation project, a website integration, or a custom workflow, you can start with Webita AI.

Repository: https://github.com/webita/blender-codex-mcp