How does Blender MCP integrate with LLM clients?

It uses a Blender add-on to open a socket and a Python MCP server to translate tool calls into Blender operators. It works with MCP-capable clients like Claude Desktop, Cursor, VS Code, and OpenCode.

Is Blender MCP free to use?

Yes, it is free and open-source under the MIT license. There is no hosted service or subscription; users only pay for the LLM used by their chosen client.

What are the security risks of using Blender MCP?

The execute_blender_code tool allows arbitrary Python execution within the Blender session, which poses a security risk if untrusted prompts are used. Users should sandbox prompts and avoid pointing it at unknown files.

Blender MCP

Give Claude hands inside Blender — an MCP server for natural-language 3D modeling.

Free· Free and open-source (MIT license). No hosted service, no subscription; you supply your own MCP-capable client (Claude Desktop, Cursor, VS Code, OpenCode) and pay only for whatever LLM the client uses.MCP ServersModel-agnostic; commonly paired with Claude (Sonnet/Opus) via Claude Desktop, but works with any MCP-capable client

Visit website →

In short

Blender MCP enables LLMs to drive Blender via natural language, allowing users to build scenes, edit materials, and import assets. It is best for technical artists and indie developers who want an AI agent that can directly manipulate 3D files rather than just provide instructions.

Best for

Blender-native technical artists, indie game devs, and 3D hobbyists who already have an MCP-capable client and want an AI agent that can genuinely build and edit scenes, not just narrate them.

Skip if

Non-technical users who want a one-click 'text to 3D' web app, studios that forbid arbitrary code execution in DCC tools, or anyone not already comfortable installing Blender add-ons and editing client JSON.

Blender MCP is an open-source Model Context Protocol server that wires Blender directly into an LLM client so a model like Claude can drive the 3D application through natural language. It consists of two pieces: a small Blender add-on that opens a socket inside the running Blender process, and a Python MCP server (launched via uvx) that translates MCP tool calls from the client into Blender operator calls and Python snippets on the other end of that socket. Once connected, the assistant can inspect the scene graph, add and transform primitives, edit materials and lighting, download assets from Poly Haven (HDRIs, textures, models) and Sketchfab, generate meshes on demand through the Hyper3D Rodin integration, capture viewport screenshots for visual feedback, and, as an escape hatch, execute arbitrary Python inside Blender. Typical workflows include roughing out a scene from a text brief ('a lantern-lit dungeon with a sleeping dragon'), staging product renders, generating props from reference images, iterating lighting and materials by voice/chat, or scripting repetitive cleanup tasks that would otherwise be Python drudgery. It is aimed at technical artists, indie game developers, VFX hobbyists, and anyone who already lives in Blender and wants an agent that can actually manipulate the file, not just describe what to do.

Editor's take

This is the reference implementation for 'LLM-driven DCC tool' and the reason a lot of people first installed an MCP client. It genuinely works, and the Poly Haven and Hyper3D hooks turn it into a small asset pipeline, not just a scripting shim. Just treat execute_blender_code like a loaded gun — sandbox your prompts and don't point it at unknown .blend files.

— The AI Tool Bible editorial team

Pros

✅ Extremely popular and actively maintained MCP server (25k+ GitHub stars) with a clear install path
✅ Real two-way integration: the LLM can inspect the scene, act, then look at a screenshot and iterate
✅ Bundled asset integrations (Poly Haven, Sketchfab, Hyper3D Rodin) let the agent pull in production-ready models and HDRIs
✅ Works with several major MCP clients (Claude Desktop, Cursor, VS Code, OpenCode), not locked to one vendor
✅ MIT-licensed and fully local — no cloud rendering, no telemetry, your .blend files never leave the machine
✅ The execute_blender_code tool is a genuine power-user escape hatch for anything the higher-level tools don't cover

Cons

⚠️ execute_blender_code runs arbitrary Python in your Blender session, which is a real security risk if you paste in untrusted prompts
⚠️ Requires a manual setup dance (install add-on, edit client JSON, run uvx server) that will trip up non-technical artists
⚠️ Quality depends entirely on the connected LLM — small or cheap models struggle to plan multi-step scene edits
⚠️ Complex operations often need to be broken into small steps or they silently fail or produce garbage geometry
⚠️ No GUI feedback inside the client beyond screenshots — long agent runs can drift from what the artist actually wanted
⚠️ Sketchfab/Hyper3D asset integrations depend on third-party accounts and API keys with their own quotas and terms

Use cases

AI-assisted 3D scene blockingGenerative prop and mesh creation via Hyper3D RodinPoly Haven HDRI and texture stagingSketchfab asset search and importMaterial and lighting iteration by chatViewport screenshot review loopsBlender Python scripting via natural languageReference-image to 3D model workflowsRapid product render setup

Frequently asked

How does Blender MCP integrate with LLM clients?: It uses a Blender add-on to open a socket and a Python MCP server to translate tool calls into Blender operators. It works with MCP-capable clients like Claude Desktop, Cursor, VS Code, and OpenCode.
What specific 3D tasks can the AI perform?: The assistant can inspect the scene graph, add and transform primitives, edit materials and lighting, download assets from Poly Haven and Sketchfab, generate meshes via Hyper3D Rodin, and capture viewport screenshots.
Is Blender MCP free to use?: Yes, it is free and open-source under the MIT license. There is no hosted service or subscription; users only pay for the LLM used by their chosen client.
Who is this tool designed for?: It is aimed at technical artists, indie game developers, and VFX hobbyists who are comfortable installing Blender add-ons and editing client JSON. It is not suitable for non-technical users seeking a one-click web app.
What are the security risks of using Blender MCP?: The execute_blender_code tool allows arbitrary Python execution within the Blender session, which poses a security risk if untrusted prompts are used. Users should sandbox prompts and avoid pointing it at unknown files.