Blender Skills
94 specialist Blender skills that turn Cursor, Claude Code, Kiro, and Codex into a senior technical art team, executing the full pipeline from blockout to engine export directly in Blender via MCP.
- Source repo
- arjun988/blender-skills
- Stars
- ★ 1.5k
- Last updated
- 3mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 46/100 · Major gaps
At a glance
- Works with
- Universal · cross-platformCodex · Claude Code
- You'll need
- Typical use
- An indie game developer needing an engine-ready hard-surface asset can request a "game-ready sci-fi rifle for Unreal Engine, 3k tris, realistic PBR" and get the full chain through export-pipeline.
- Main limitation
- Requires a running Blender plus the third-party BlenderMCP addon; if the MCP connection fails, the skills cannot execute.
- Source review
- 46/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
Blender Skills is an MIT-licensed collection of 94 specialized skills with 175+ reference files, stored in .claude/skills/ and mirrored to .cursor/skills/. Its orchestrator skill, blender-director, analyzes a request and automatically routes it through a skill chain such as hard-surface modeling, UVs, materials, optimization, and export. Every skill operates on a running Blender instance through the Blender MCP (BlenderMCP addon) rather than narrating UI steps. Coverage spans modeling, sculpting/retopology, technical tasks, pipeline, art styles (including horror packs), worlds/moods, and game genre direction. The repository ships per-client installation steps and MCP config files for Cursor, Claude Code, Kiro, and Codex. Per the README, the Anthropic marketplace listing is under review, so installation is currently direct from GitHub.
Once installed, your coding agent reads the skill files under .claude/skills/ (or .cursor/skills/ for Cursor); blender-director parses the request and dispatches sub-skills in order. Skills execute real operations inside Blender via Blender MCP tools: modeling (blender-modeler, hard-surface, environment-artist, etc.), sculpting and retopology, UVs and materials (uv-workflow, materials, texture-workflow), technical tasks (geometry-nodes, lighting, rigging, animation), and pipeline stages (asset-optimization, lod-pipeline, collision-proxy, unity-export, unreal-export, godot-export, qa-review). Reference image matching is supported: reference-image-match performs camera matching, builds geometry in tiers, and loops on screenshot comparison. Workflows follow production standards for naming conventions and polycount budgets, and validate via checklist gates before export.
- An indie game developer needing an engine-ready hard-surface asset can request a "game-ready sci-fi rifle for Unreal Engine, 3k tris, realistic PBR" and get the full chain through export-pipeline.
- An environment artist building modular kits or horror scenes can combine horror-style, environment-artist, and lighting skills.
- A character artist with a creature sculpt can run retopology, uv-workflow, texture-workflow, and rigging against a specified triangle budget for game animation.
- Concept-driven teams can attach a photo or concept art so reference-image-match performs camera matching and a screenshot-comparison loop to reproduce the reference.
- Teams needing a specific look can invoke style skills (psx-horror-style, lowpoly-style, anime-style) paired with the matching export pipeline.
How do you install or deploy this agent?
Prerequisites: install uv; install the BlenderMCP addon (github.com/ahujasid/blender-mcp) in Blender; in Blender's N sidebar → BlenderMCP → Connect to Claude.
Claude Code:
git clone https://github.com/arjun988/blender-skills.git
cp -r blender-skills/.claude/skills /path/to/your/project/.claude/skills/Cursor (Windows):
git clone https://github.com/arjun988/blender-skills.git
Copy-Item -Recurse -Force "blender-skills\.claude\skills\*" ".cursor\skills\"Kiro:
cp -r blender-skills/.claude/skills/* /path/to/kiro/skills/Codex:
cp -r blender-skills/.claude/skills/* /path/to/codex/skills/
# or: ln -s /path/to/blender-skills/.claude/skills /path/to/codex/skills/blender-skillsThe repo ships .cursor/mcp.json and .mcp.json as the Blender MCP configs for the respective clients. The README notes the Anthropic marketplace listing is under review; install directly from GitHub for now.
How do you use this agent?
- Start Blender and connect via BlenderMCP in the N-panel sidebar. 2. Restart your agent (Cursor / Claude Code / Kiro / Codex). 3. Issue a natural-language request, e.g.:
Build a game-ready sci-fi rifle for Unreal Engine, 3k tris, realistic PBRblender-director routes to hard-surface → uv-workflow → materials → asset-optimization → export-pipeline and executes directly in Blender. You can also attach a reference image and ask it to "analyze first, then build in Blender"; the skill camera-matches and loops screenshot comparisons. The full skill index is in SKILLS_GUIDE.md; MCP setup in docs/BLENDER_MCP_SETUP.md.
What are this agent's strengths and limitations?
- 94 skills plus 175+ reference files cover the entire production pipeline from modeling to unity/unreal/godot export, not a single-purpose script.
- Executes via Blender MCP with no narrated UI steps, and gates every delivery on asset-optimization and checklist validation.
- The same skill pack adapts to Cursor, Claude Code, Kiro, and Codex, with per-client directory structures and MCP config files included.
- Requires a running Blender plus the third-party BlenderMCP addon; if the MCP connection fails, the skills cannot execute.
- Cursor users must manually copy skill folders into .cursor/skills/, so multi-project teams maintain sync themselves.
- The Anthropic marketplace listing is still under review; there is no one-click install channel today.
- Output quality depends on the underlying model's Blender operation ability; the repository provides no benchmarks or quality evidence.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Blender Skills This agent | 46 · Major gaps | — | ★ 1.5k | 3mo ago | TypeScript | Codex · Claude Code |
| UI Craft | 77 · Good | CLIFree | ★ 407 | 10d ago | JavaScript | Codex · Claude Code |
| oh-my-design | 76 · Good | CLIFree | ★ 549 | 4d ago | JavaScript | Codex · Claude Code |
| Talk to Figma MCP | 48 · Major gaps | MCP serverFree | ★ 7.1k | 3mo ago | JavaScript | — |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows the pack drives an external application (Blender) directly via MCP; the README states no permission scope, confirmation gates, or destructive-action safeguards. The actual skill files are not provided, so least_privilege and user_confirmation inside skill instructions cannot be verified — scored 1 as assertion-only. No mention of sensitive data handling. Dependencies point to external uv and the third-party BlenderMCP addon with no pinning or security notes. No rollback/undo mechanism described. Attribution is present (credit to jeffallan/claude-skills), scored 2.
README is internally consistent (94 skills, structure, client matrix align), but the 94 skill directories themselves are not provided for verification; availability of the external BlenderMCP addon and uv has no fallback documented; no evidence of error handling or failure messages.
Audience (Blender artists/tech artists) and scenarios (hard-surface, character, horror, engine export) are well described, scored 2; four-client environment coverage is clear. Capability boundaries are thin — only one philosophy line about 'never narrates UI steps', nothing on degraded behavior when MCP is disconnected; trigger precision rests on the blender-director routing description whose content is not supplied.
Good documentation architecture: QUICKSTART, SKILLS_GUIDE, MCP setup, CONTRIBUTING, CHANGELOG, references all indexed; install notes cover four clients including a Windows PowerShell variant, scored 2; naming consistency visible in README tables but unverifiable across all 94; rich workflow examples; known limitations nearly absent (only 'marketplace under review'), scored 1; complete MIT LICENSE scored 3; version badge 1.3.0 + version.json + CHANGELOG referenced but contents not provided; maintainer identity and update commitment unclear, publisher unverified.
Output positioning is clear (game-ready assets, polycount budgets, export validation gates) with plausible marginal value for the target user; however all effectiveness claims are unverified, and the cost side (external addon + uv + four-step setup) is not weighed against benefit.
Core claims (94 skills, 175+ reference files, 'executes directly in Blender') cannot be traced in the supplied files; 'Marketplace listing under review by Anthropic' is an unverifiable claim with no fact/inference separation; only README and LICENSE are available, so no cross-source corroboration is possible.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- This review is based only on README and LICENSE; the actual content of the 94 skills (permissions, confirmations, error handling) is entirely unverified — inspect the instruction files under .claude/skills/ yourself.
- The agent modifies a running Blender session via MCP with no documented confirmation or undo mechanism; trial in non-critical scenes first and avoid destructive operations on important .blend files.
- Depends on the third-party BlenderMCP addon (ahujasid/blender-mcp) and uv with no version pinning; assess their security independently before use.
- Numeric claims (94 skills, 175+ reference files) are unverified; the 'Anthropic marketplace under review' statement is unverifiable.