Skip to content

MCP

What is MCP? Connect an AI assistant to Blender

Learn what Model Context Protocol does, connect an AI assistant to Blender, and verify the connection by listing objects in a practice scene.

Jump to setup
A compact hub connecting several distinct cables at a workstation

MCP gives an AI app a standard way to use tools from other software. In Blender, that means an assistant can look at the objects in your open scene. You can test the connection by asking it to list those objects and comparing its answer with Blender’s object list.

Step-by-step setup

Connect an AI assistant to Blender with MCP

What you need Blender, the Codex CLI or another host that supports local MCP servers, and uv with its uvx command. Start in a new practice scene. This example uses the community MCP for Blender add-on.

  1. Your prompt
  2. AI app
  3. MCP server
  4. Blender add-on
  5. Open scene
  1. Understand the connection

    The AI app sends a request to the MCP server. The server talks to the Blender add-on. The add-on reads the open scene and returns the result to the AI app.

  2. Install the server and add-on

    Install uv using its official guide. In a terminal, run the sample commands. The first installs the Blender add-on, the second registers the MCP server in Codex, and the third checks that it is listed. If Blender is already connected through a plugin, use that connection instead of adding a second server.

  3. Start the Blender connection

    In Blender open Edit → Preferences → Add-ons and enable MCP for Blender. In the 3D Viewport press N, open the MCP for Blender tab and start the server if the panel says it is stopped. Keep Blender open and start a new Codex session on the same computer.

  4. Ask for one read-only action

    In a fresh General scene, rename Cube to PracticeCube in the Outliner, which is the panel listing objects. In chat ask: “Use Blender MCP’s get_scene_info to list the objects in the open scene. Do not edit or save anything.”

Terminal commands for Codex CLI

uvx mcp-for-blender install-addon
codex mcp add blender -- uvx mcp-for-blender
codex mcp list

How to tell it worked

The assistant reports PracticeCube, Camera and Light from a tool call, and those names match the Outliner. If it cannot connect, check that the add-on is running in Blender and restart the AI app after registering the server.

The community add-on can also edit Blender scenes and run Python. Asking for a read-only action does not remove those permissions. Use the companion guide to allow only scene inspection in Codex.

Limit the Blender connection to scene inspection

Meet the host, client and server

The host is the AI application that coordinates the interaction. An MCP client in the host maintains the connection to a server. The server exposes capabilities, such as looking up a course, and returns results when it receives a supported request. The underlying data can come from a file, service or database.

Understand tools, resources and prompts

A tool is a callable capability, such as searching approved course descriptions. A resource can provide useful context, such as a course outline document. A prompt can offer a reusable template, such as a checklist for summarizing a syllabus. What you can access or see depends on the server and the host.

A connected-node illustration with a central hub and separate endpoints
Think of MCP as a common way for an AI app to discover and request capabilities from connected servers.

Separate connection from decision making

MCP describes how clients and servers communicate. The workflow still needs to decide which tools are appropriate, how results are checked and when to ask a person for approval. A server that can read a catalogue does not need the ability to change it.

Start with one purpose and one source

Choose a small, owned dataset such as a catalogue made for practice. Expose only the lookup that the project needs. When a result is missing, have the assistant explain that it could not find the answer and suggest where a person can confirm it.

Try it yourself

Draw a line from an AI assistant to a client, a course catalogue server and the catalogue itself. Label each step for a question about syllabus topics. Draw the return path and note where the source reference should appear.

Common questions

Is MCP an AI model?

No. It is a protocol for exchanging context and capabilities. A host can use it with models and workflows of its choice.

Does every server offer the same tools?

No. Each server exposes capabilities for its intended use. Some only provide access to data.

Sources and further reading