Blender MCP

- Scene performance analysis— Ask your AI to analyze the open Blender scene and identify objects with high polygon counts relative to their on-screen size, like the classroom demo outlier detection.
- Data-block renaming— Have the assistant fix typos or suggest descriptive names for all data-blocks in the current file, applying changes only after your approval.
- Natural language scene queries— Ask questions about object relationships, such as "Which objects are using the material 'pebbles'?" or identify the highest poly-count object in the scene.
- Geometry Nodes documentation— Request an explanation of the main geometry nodes setup, with inline frame documentation and a text data-block summarizing the analysis.
- Scene debugging— Get help troubleshooting issues like non-uniformly scaled meshes, bad normals, or objects not deforming correctly with armatures.

A lightweight MCP (Model Context Protocol) server for Blender. It offers a natural language interface with Blender’s Python API, improving access to documentation, and allowing users to explore and understand complex setups.

For technical details and documentation about the architecture check thesource code.

[!warning] WarningSecurity Warning
The MCP server will execute LLM generated code in Blender without any guards in place to protect your data from removal or being sent to a remote location. To keep your data safe it is recommended to use a virtual machine, or a system without access to sensitive information.

Blender does not have any built-in functionality for connecting to LLMs.

In order for Blender to connect with LLMs,three external tools must be manually downloaded, installed, and run.

In order to interact with your Blender session you need to install a specific add-on for the MCP Server integration.

If you drag & drop into Blenderyou will need to do it twice. First to add the Blender Lab repository, secondly to install the add-on itself. This method allows you to receive update notifications whenever a new version of this add-on is available.

MCP Servers follow a well defined standard and are compatible with a myriad of clients. Please follow theLLama.cppdocumentation, or install a LLM client of your preference.

- MCP Bundle: For newer clients that support.mcpbfiles, download thelatest packagefrom the release page.Llama.cpp doesn’t support this yet.
- MCP Server: To install from the source code check theLlama.cpp documentationor thesetup instructions.

MCP can be used to programatically analyze the scene for performance bottomnecks. Take for instance theClassroom demo file:

This scene is one of theBlender benchmarkfiles. The reason those files are in the benchmark is because they are a good representation of real production files created by artists. And it is no surprise that it can contain objects prone to be optimized.

[!info] InfoHow reliable are those results?

The initial analysis returned by the LLM only considered the modifiers which influenced the viewport. Thecoat 1object has a Solidify modifier which doubles its poly count, making it an even more outlier. And although this scene had no Simplify enabled, this settings would have also affected the final analysis.

Those are some of the other tested use cases. You will need to paste the entire “Prompt” content for them to work. As a start point you should use the corresponding demo file. The sucess of the operation will depend on the model used.

- Translate all the data-blocks from French to English.
- My mesh is not being deformed by my armature, how can I fix that?
- Blender is running out of memory rendering this scene, how can I optimize it?
- This mesh has strange shading artifacts, how can I you fix that?
- Find objects that have meshes with bad normals.
- Check my scene for non-uniformally transformed mesh objects.
- The video I exported doesn’t play in my web browsers, which settings should I change?
- Which sculpt brushes should I use for creating and detailing tree branches?
- Set up compositing nodes to save my image both as SDR and HDR.
- Verify this checklist:: meshes must be manifold, all objects must have materials, naming must follow convention, no absolute paths, … Data-block renaming: fix typos

GRP-rocks→GRP-pebbles(the scene contains only pebble objects, not rocks).

LGT-Lights→LGT-lights(inconsistent casing).

Compositing Nodetree→Compositing Node Tree(Nodetree is a compound word error).

- Camera→CAM-main
- Area→LGT-sun-key
- Area.001→LGT-sun-fill
- Area.002→LGT-area-rim
- GEO-pebble.001→GEO-pebble-B
- GEO-pebble.002→GEO-pebble-C
- GEO-pebble.003→GEO-pebble-D
- GEO-pebble.004→GEO-pebble-E
- GEO-pebble.005→GEO-pebble-F
- GEO-pebble.006→GEO-pebble-G

- Area→LGT-sun-key
- Area.001→LGT-sun-fill
- Area.002→LGT-area-rim

- Texture→TEX-ground-clouds
- Texture.001→TEX-ground-image

- World→WLD-environment
- Scene→SCN-mainGeometry Nodes Documentation

Documentation frames added to the geometry nodes setup.

GEOMETRY NODES ANALYSIS -- "Pebble Scattering Nodes" Applied to object: GEO-ground | Modifier: Pebbles ======================================================== OVERVIEW -------- This node group scatters three size categories of pebble meshes across the faces of a ground mesh. Each category uses its own density mask and a global Factor multiplier so that coverage can be art-directed per-layer without touching the raw mask textures. The three resulting instance layers are merged with the original ground geometry and sent to the Group Output. INPUTS (Group Interface) ------------------------ Geometry -- The ground mesh to scatter pebbles on. Large Pebbles Mask -- Float mask (e.g. vertex colour / texture) that defines WHERE large pebbles may appear. Medium Pebbles Mask-- Float mask for medium pebble placement regions. Small Pebbles Mask -- Float mask for small pebble placement regions. Factor -- Global density multiplier shared by all three lanes (appears three times in the interface, one per lane). NODE GRAPH -- LANE BY LANE -------------------------- +- LARGE PEBBLES LANE -----------------------------------------------------+ | Source object : GEO-pebble | | Group Input.001 | | +- Geometry ------------------------> Distribute Points on Faces | | +- Large Pebbles Mask ------------> Distribute Points on Faces | | | (Density Factor socket) | | +- Factor --> Math.003 (x) -------> Distribute Points on Faces | | (Density Max socket) | | Distribute Points on Faces ----------> Instance on Points | | Object Info (GEO-pebble) -----------> Instance on Points (Instance) | | Random Rotation.001 [-pi, +pi] -----> Instance on Points (Rotation) | | Random Value [0.25 - 0.60] -------> Instance on Points (Scale) | | Instance on Points ------------------> Join Geometry.003 | +--------------------------------------------------------------------------+ +- MEDIUM PEBBLES LANE ----------------------------------------------------+ | Source object : GEO-pebble.004 | | Group Input.002 | | +- Geometry ------------------------> Distribute Points on Faces.001 | | +- Medium Pebbles Mask -----------> Math.004 (x) --> Math.005 (x) | | +- Factor ------------------------> Math.005 (x) | | Math.005 output --------------> Distribute Points on Faces.001 | | (Density socket) | | Distribute Points on Faces.001 -----> Instance on Points.001 | | Object Info.001 (GEO-pebble.004) --> Instance on Points.001 | | Random Rotation.002 [-pi, +pi] ----> Instance on Points.001 | | Random Value.002 [0.25 - 0.45] -----> Instance on Points.001 | | Instance on Points.001 -------------> Join Geometry.003 | +--------------------------------------------------------------------------+ +- SMALL PEBBLES LANE -----------------------------------------------------+ | Source object : GEO-pebble.002 | | Group Input.003 | | +- Geometry ------------------------> Distribute Points on Faces.002 | | +- Small Pebbles Mask -----------> Math.006 (x) --> Math.007 (x) | | +- Factor ------------------------> Math.006 (x) | | Math.007 output --------------> Distribute Points on Faces.002 | | (Density socket) | | Distribute Points on Faces.002 -----> Instance on Points.002 | | Object Info.002 (GEO-pebble.002) --> Instance on Points.002 | | Random Rotation [-pi, +pi] ----> Instance on Points.002 | | Random Value.003 [0.10 - 0.35] -----> Instance on Points.002 | | Instance on Points.002 -------------> Join Geometry.003 | +--------------------------------------------------------------------------+ +- MERGE & OUTPUT ---------------------------------------------------------+ | Group Input.004 | | +- Geometry (pass-through) --------> Join Geometry.003 | | Join Geometry.003 | | (inputs: large instances + medium instances + | | small instances + original ground geo) | | +- Geometry -----------------------> Group Output | +--------------------------------------------------------------------------+ SCALE RANGES (uniform, per lane) --------------------------------- Large pebbles : 0.25 - 0.60 Medium pebbles : 0.25 - 0.45 Small pebbles : 0.10 - 0.35 ROTATION (all lanes) --------------------- All three axes randomised independently over [-pi, +pi], giving each pebble instance a fully random orientation. DENSITY CONTROL PATTERN (Medium & Small lanes) ----------------------------------------------- The mask value and the Factor are first multiplied together (Math.004 / Math.006), then that product is multiplied again by a second value (Math.005 / Math.007) before being fed into the Density socket. This two-stage multiply gives a non-linear response curve, making the density fall off more aggressively near the mask edges. The Large lane uses a different (single-stage) approach: the Density Factor socket receives the mask directly, and the Factor is only used to scale Density Max via Math.003. NOTES & SUGGESTIONS --------------------  The node group has no Seed input exposed; adding one would allow re-randomising all three layers simultaneously without touching individual nodes.  The Factor input currently appears three times (once per lane). Merging them into a single shared socket would simplify the modifier panel. * Consider labelling the unlabelled Math nodes (Math.003-.007) and Random Value nodes to aid future maintenance.

You're agent can Chain 60+ AI image and video models on one workflow canvas

MCP Server for dynamically generating infographics using Napkin.AI

The Lottie Creator MCP brings your AI assistant directly into your animation workflow — giving it full access to LottieFiles Creator so it can build and edit Lottie animations on your behalf through natural language.

AI Video, Image & Audio Generation with over 150 models

Rive MCP let AI handle repetitive tasks, like creating complex View Models, State Machines with hundreds of states/layers, Layouts, Shapes, and more.

Create, edit, and export SVGator animations (SVG, Lottie, GIF, video) from your AI assistant, and keep an editable project you can open in the editor anytime.

Convert photos to LEGO-style brick mosaic previews with AI-powered color quantization. MCP server for Claude, Cursor, and other AI agents.

Changeimageto - Free Image Editing tools

16 tools to remove backgrounds, blur or grayscale backgrounds, change colors, convert formats (PNG/JPEG/WebP/…), upscale, denoise/enhance, run OCR, export PDF, edit text in images (Gemini), clean watermarks, inpaint masked regions, and run bulk resize / convert / quality checks.

Edit your Church website with ChatGPT or Claude

Hosted remote MCP server for AI image and video generation from one Streamable HTTP endpoint.