What Are the Best Blender Skills for Claude Code?

Blender is scriptable end to end through its Python API, which makes it one of the best 3D tools to drive from Claude Code, yet most developers still click through the GUI for every export. The skills below let the agent inspect scenes, fix rigs, and ship web-ready assets, starting with blender-motion-state-inspection.

Key takeaways

• The most-installed Blender skill in the Skillselion catalog is blender-motion-state-inspection at 4,580 installs (skills.sh registry), from the affaan-m/ecc collection (248,575 GitHub stars).
• "blender mcp server" is a rising search, up 60% in the last 90 days (Google Trends).
• blender-web-pipeline (2,397 installs) covers glTF export and PBR optimization for Three.js and Babylon.js.
• blender-mcp (1,738 installs) connects the agent to PolyHaven so models, HDRIs, and textures import automatically.
• The catalog lists these alongside 59,535 other skills; the design view lives in the animation hub and the wider frontend design category.

Why drive Blender from Claude Code at all?

Because screenshots alone cannot tell an agent what is wrong with a 3D scene, and Blender's bpy API can. A rendered frame shows a character looks twisted; only structured data shows which bone is mirrored or where a foot loses ground contact. Skills that read scene facts let Claude Code debug 3D the way it debugs code: with evidence instead of guesses.

What is the best Blender skill for Claude Code?

blender-motion-state-inspection is the top pick at 4,580 installs (skills.sh registry), part of affaan-m/ecc with 248,575 GitHub stars. It extracts bone hierarchies, bounding boxes, foot-ground contacts, and facing vectors from .blend files, then compares them against rendered evidence to catch twisted, mirrored, flattened, or foot-sliding rigs that screenshots cannot diagnose.

npx skills add https://github.com/affaan-m/ecc

A close second for shell-first workflows is cli-anything-blender (724 installs, hkuds/cli-anything, 48,986 stars), which edits and renders 3D scenes from JSON scene descriptions without ever opening the GUI. A separate skill named blender from sfkislev/flue (2,536 installs) takes the bridge approach, piping Python into Blender's own bpy runtime and returning structured JSON.

How do you get Blender models onto the web?

Export through blender-web-pipeline, which packages 2,397 installs worth of glTF export workflows, pre-configured templates, and optimized PBR shaders for Three.js and Babylon.js targets (skills.sh registry). glTF is the Khronos Group's royalty-free standard for runtime 3D assets, so the output drops straight into a web scene. From there, our Three.js guide covers the rendering side.

How does blender-mcp fit in?

blender-mcp (1,738 installs, skills.sh registry) gives the agent MCP tools for scene inspection, Python scripting, glTF export, and materials, plus direct search and import from PolyHaven's asset library. Use the structured tools for quick checks and execute_python for bulk operations like hierarchy traversal or animation baking. Blender itself stays the single source of truth for the scene.

What does an agent-driven Blender workflow look like?

A typical session runs inspect, fix, export, verify, all without opening the Blender window. The agent uses blender-mcp's get_scene_info and screenshot tools to establish what exists, then blender-motion-state-inspection to pull bone hierarchies and contact data when something looks wrong in a render. Fixes go through execute_python or the shell bridge as small bpy scripts the agent writes and runs, and each one returns structured JSON so the agent can confirm the change landed. When the scene passes inspection, blender-web-pipeline exports optimized glTF, and the agent loads it in a Three.js test page to verify materials survived the trip. The pattern that matters: every step produces machine-readable evidence, which is why these four skills together carry more than 9,400 installs (skills.sh registry, September 2026) among developers shipping 3D on the web.

A complete starter stack

• blender-motion-state-inspection - rig and animation diagnosis (4,580 installs)
• blender-web-pipeline - glTF export for the web (2,397 installs)
• blender-mcp - MCP control plus PolyHaven assets (1,738 installs)
• cli-anything-blender - GUI-free scene editing and rendering (724 installs)

Common pitfalls

• Diagnosing rigs from screenshots only: retargeting errors like mirrored bones need structured scene facts, not another render.
• Exporting unoptimized assets: skipping the PBR and mesh optimization step produces glTF files too heavy for the web.
• Mixing bridge styles: pick MCP tools or a shell bridge per project; running both against one Blender instance causes state confusion.

Give the agent real scene data and Blender stops being a manual bottleneck; start with the ranked list in the animation hub, then widen out to the best skills for frontend design.

Common questions