Delv
CommunityActive· 8d4.3by Siddharth Ahuja

Blender

Siddharth Ahuja's Blender MCP. Lets Claude drive Blender for 3D modelling, scene composition, render automation.

C
Safety & Trust

Delv Safety Grade: C

Score 58/100 · assessed 2026-04-28

Maintainer45
Permissions35
Supply chain72
Transparency68
Incidents100

Blender MCP is a community server by solo developer Siddharth Ahuja that bridges Claude to Blender's Python API for 3D modelling and rendering automation. The server executes arbitrary Python code inside Blender, which means it has desktop application control and can run shell commands through Blender's scripting environment. It's distributed via PyPI as a proper package with uvx installation, which is good for supply chain hygiene. The repository is open source with reasonable documentation, though it's a personal project with single-maintainer risk. No security incidents are known, but the permission model is inherently broad: controlling Blender means filesystem access for saving renders, network access for addon downloads, and execution of user-supplied Python. Suitable for trusted local workflows, not for untrusted prompts or shared environments. The technical implementation is clean, but the attack surface is substantial.

Lethal Trifecta (prompt-injection exposure)

CLEAR
Private dataNo
Reads secrets, credentials, private files
Untrusted inputNo
Ingests web pages, PRs, issues, emails
External commsNo
Can send data outbound

Local 3D. No I/O.

Green flags

  • Distributed via PyPI with standard package management (uvx)
  • Open source repository with clear documentation and examples
  • No environment variables or secrets required for basic operation
  • No known security incidents or malicious versions
  • Specific, well-scoped use case (Blender automation only)

Red flags

  • Executes arbitrary Python code in Blender with full scripting privileges
  • Solo maintainer project with limited bus factor and review
  • Desktop application control enables filesystem and potential shell access
  • No sandboxing of Python execution within Blender environment
  • Could be used to exfiltrate data via render output or addon installation

Permissions requested

Desktop controlWrite filesRead filesShell executeOutbound network
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

uvx blender-mcp

Review

This MCP server turns Blender into a conversational 3D workshop. You describe what you want, Claude writes the Python commands, and Blender executes them. It's not a prompt-to-mesh miracle, it's an API bridge that lets you script Blender without leaving your chat window. I've used it to rough out scene layouts from text descriptions and to automate camera setups for product renders. The sweet spot is procedural work: generating arrays of objects, tweaking materials across multiple meshes, or queuing batch renders with different lighting conditions. You still need to know what a UV map is and how Blender's node system works, but you can skip the GUI hunting. The server exposes Blender's Python API, so anything you could script manually, you can now ask Claude to script for you. Quirks: Blender must be running and listening before you start chatting. The server doesn't launch Blender for you. If your prompt is vague, Claude will make sensible guesses, but you'll often need a second pass to refine positions, scales, or materials. Complex geometry, like organic sculpting or hard-surface detailing, still wants manual intervention. This isn't a replacement for modelling skills, it's a faster way to execute ideas you already understand. The procedural modelling workflow is where it shines. I asked it to create a grid of cubes with randomised heights and colours, then adjust the palette based on a hex code list. That would've been tedious in the GUI, but took two prompts here. Same for setting up a turntable render: specify the object, frame count, and output folder, and it writes the animation and render queue. Who shouldn't bother: if you're new to Blender, this won't teach you the fundamentals. If you only model organically or sculpt, the automation gains are minimal. But if you're a technical artist, a procedural nerd, or someone who renders the same scene with different variables, this is a genuine time-saver. It's also useful for prototyping ideas before committing to manual refinement.
Verdict

Install this if you already script Blender or wish you did. It's a clean bridge between natural language and Blender's Python API, best for procedural tasks and render automation. Skip it if you're learning 3D from scratch or rarely touch parametric workflows.

Good at

  • Exposes Blender's full Python API through natural language, so anything scriptable is now conversational.
  • Excellent for procedural modelling, batch operations, and render queue automation.
  • No need to context-switch between chat and Blender's scripting console.
  • Works well for iterative refinement, you can tweak parameters in follow-up prompts without starting over.

Watch out

  • Requires Blender to be running and configured before the MCP session starts, no automatic launch.
  • Vague prompts lead to generic results, you need to know what you're asking for in Blender terms.
  • Not useful for organic modelling, sculpting, or tasks that rely on visual feedback loops.
  • Limited to Claude Desktop and Cursor unless you manually configure other hosts.

Use cases

  • Generating a 3D scene from a prompt
  • Procedural modelling with iteration
  • Setting up a render queue from chat
  • Bridging concept art to 3D

Getting started

1. Install via `uvx blender-mcp` in your terminal. 2. Add the server to your Claude Desktop config under `mcpServers` with the command pointing to the installed binary. 3. Launch Blender and ensure it's running with the Python API accessible (check the scripting workspace to confirm). 4. In Claude Desktop, ask it to create a simple object like a cube or sphere to verify the connection. 5. Watch out: Blender must be open before you start the conversation, or the server will fail silently.

Works with

Claude DesktopCursor

Similar MCPs