Blender MCP:用自然语言控制 Blender 3D,接任意 LLM,26K stars 社区插件

blender-mcp-control-blender-3d-any-llm-mcp-claude-code

Tech-News #Blender#MCP#3D建模#Claude#LLM#AI工具#开源#模型上下文协议
更新于
🇨🇳 中文

by Mycelium Protocol


GitHub:ahujasid/blender-mcp
官网:blendermcp.org
PyPI:blender-mcp
许可证:MIT
语言:Python
Stars:26,126 · Forks:2,484
作者:Siddharth(@sidahuj)


一、一句话

在 Claude 里说「创建一个地下城场景,龙守着一罐金币」,然后 Blender 里真的出现了。

这就是 Blender MCP 做的事:把自然语言指令通过 MCP 协议传给 Blender,由 LLM 控制 3D 软件执行操作。


二、架构:两个组件

Claude / 任意 LLM
       ↕  MCP 协议
  MCP Server(Python)
       ↕  TCP Socket / JSON
  Blender Addon(addon.py)
       ↕  Blender Python API
  Blender 3D 引擎

Blender Addon(addon.py):在 Blender 内部创建一个 socket 服务器,监听来自 MCP Server 的命令,调用 Blender Python API 执行,返回结果。

MCP Server(src/blender_mcp/server.py):实现 Model Context Protocol,对 LLM 暴露工具集,把 LLM 的调用翻译成 TCP 消息发给 Blender。

通信协议:JSON over TCP,默认端口 9876。指令格式:{type, params},响应格式:{status, result/message}。


三、安装(三步)

步骤 1:安装 uv

# macOS
brew install uv

# Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

注意:用官方安装器,不要用 pip install uv——后者不一定创建 uvx 命令,会导致客户端找不到它。

步骤 2:配置 MCP 客户端

Claude Code(一条命令):

claude mcp add blender uvx blender-mcp

Claude Desktop(claude_desktop_config.json):

{
    "mcpServers": {
        "blender": {
            "command": "uvx",
            "args": ["blender-mcp"]
        }
    }
}

多版本 Python / conda 环境(避免冲突,锁定 3.11):

{
    "mcpServers": {
        "blender": {
            "command": "uvx",
            "args": ["--python", "3.11", "blender-mcp"],
            "env": { "UV_PYTHON_PREFERENCE": "only-managed" }
        }
    }
}

Cursor / VS Code / OpenCode 也支持,详见 README 的 MCP Client Setup 章节(有一键安装按钮)。

步骤 3:安装 Blender 插件

uvx blender-mcp install-addon

然后在 Blender:Edit → Preferences → Add-ons → 启用 Interface: Blender MCP。

步骤 4:连接

在 Blender 3D 视口按 N → 找到 BlenderMCP 标签页 → 点 Start MCP Server。完成。


四、能力矩阵

能力说明
3D 对象操控创建、修改、删除几何体,设置位置/旋转/缩放
材质控制应用和修改材质、颜色、纹理
场景信息获取当前场景的详细状态(对象列表、层级关系等)
Python 执行直接在 Blender 里运行任意 Python 代码
Poly Haven下载免费资产、纹理和 HDRI 环境贴图
Sketchfab搜索并下载 3D 模型(需要 API Key)
Hyper3D RodinAI 生成 3D 模型(需要 API Key)
Hunyuan3D腾讯混元 AI 3D 生成(需要 SecretId/Key)

五、示例指令

这些是可以直接对 Claude 说的话:

"Create a low poly scene in a dungeon, with a dragon guarding a pot of gold"
"Create a beach vibe using HDRIs, textures, and models like rocks and vegetation from Poly Haven"
"Generate a 3D model of a garden gnome through Hyper3D"
"Make this car red and metallic"
"Create a sphere and place it above the cube"
"Make the lighting like a studio"
"Point the camera at the scene, and make it isometric"

也可以给一张参考图,让 Claude 在 Blender 里重建对应场景。


六、持久化凭据

API Key 可以存在 Blender 插件偏好设置里(Edit → Preferences → Add-ons → Blender MCP),在 Blender 重启后保留:

  • Sketchfab API Key
  • Hyper3D API Key
  • Hunyuan3D SecretId / SecretKey

CI/无头环境也可以用环境变量注入:BLENDERMCP_SKETCHFAB_API_KEY、BLENDERMCP_HYPER3D_API_KEY 等。


七、注意事项

安全:execute_blender_code 工具允许在 Blender 里执行任意 Python 代码。用之前先保存工作文件。

遥测:默认开启匿名使用数据收集。关闭方式:

"env": { "DISABLE_TELEMETRY": "true" }

或在 Blender → Add-on 偏好设置里取消勾选。

单实例:同时只能运行一个 MCP 服务器(Claude Desktop 或 Cursor 选一个),不要同时开两个。

超时问题:复杂操作拆成小步骤。第一条命令有时不通,从第二条开始正常——这是已知行为。


八、背景

Blender MCP 由 Siddharth(@sidahuj)维护,社区驱动,MIT 协议。26K stars,生态已经在跑:有人用它从截图重建 3D 场景,有人对接 Hyper3D 批量生成游戏资产,有人在 CI 里用无头模式自动化 3D 渲染流水线。

MCP 作为协议的价值在这里体现得很直接:Blender 本来有完整的 Python API,MCP 把这个 API 暴露给了任意 LLM,不需要改 Blender 本体,也不需要针对每个 LLM 写适配层。


Mycelium Protocol — 追踪 AI 系统的底层演化


关于 Mycelium

菌丝协议。持续追踪 AI 工具、系统和实验的内容节点。


🇬🇧 English

Blender MCP: Control Blender 3D with Any LLM, 26K Stars

by Mycelium Protocol


GitHub: ahujasid/blender-mcp
Site: blendermcp.org
PyPI: blender-mcp
License: MIT
Language: Python
Stars: 26,126 · Forks: 2,484
Author: Siddharth (@sidahuj)


The One-Liner

Tell Claude “Create a dungeon scene with a dragon guarding a pot of gold” — and Blender builds it.

Blender MCP routes natural language through the Model Context Protocol into Blender’s Python API, letting any LLM control a 3D application in real time.


Architecture: Two Components

Claude / any LLM
       ↕  MCP protocol
  MCP Server (Python)
       ↕  TCP socket / JSON
  Blender Addon (addon.py)
       ↕  Blender Python API
  Blender 3D engine

Blender Addon (addon.py): creates a socket server inside Blender that listens for commands from the MCP Server, calls the Blender Python API, and returns results.

MCP Server (src/blender_mcp/server.py): implements MCP, exposes a tool set to the LLM, and translates LLM tool calls into TCP messages to Blender.

Protocol: JSON over TCP, default port 9876. Commands: {type, params}. Responses: {status, result/message}.


Install (Three Steps)

Step 1: Install uv

# macOS
brew install uv

# Linux
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

Use the official installer — not pip install uv. The pip route may not create the uvx command that MCP clients need.

Step 2: Configure your MCP client

Claude Code (one command):

claude mcp add blender uvx blender-mcp

Claude Desktop (claude_desktop_config.json):

{
    "mcpServers": {
        "blender": {
            "command": "uvx",
            "args": ["blender-mcp"]
        }
    }
}

For conda / pyenv environments, pin Python 3.11 to avoid interpreter conflicts:

{
    "mcpServers": {
        "blender": {
            "command": "uvx",
            "args": ["--python", "3.11", "blender-mcp"],
            "env": { "UV_PYTHON_PREFERENCE": "only-managed" }
        }
    }
}

Step 3: Install the Blender addon

uvx blender-mcp install-addon

In Blender: Edit → Preferences → Add-ons → enable Interface: Blender MCP.

Press N in the 3D viewport → BlenderMCP tab → Start MCP Server. Done.


Capabilities

FeatureDescription
3D object manipulationCreate, modify, delete geometry; set position/rotation/scale
Material controlApply and modify materials, colors, textures
Scene inspectionGet detailed state of the current scene (objects, hierarchy, etc.)
Python executionRun arbitrary Python code directly in Blender
Poly HavenDownload free assets, textures, and HDRI environment maps
SketchfabSearch and download 3D models (API key required)
Hyper3D RodinAI-generated 3D models (API key required)
Hunyuan3DTencent Hunyuan AI 3D generation (SecretId/Key required)

Example Prompts

"Create a low poly scene in a dungeon, with a dragon guarding a pot of gold"
"Create a beach vibe using HDRIs, textures, and models like rocks and vegetation from Poly Haven"
"Generate a 3D model of a garden gnome through Hyper3D"
"Make this car red and metallic"
"Create a sphere and place it above the cube"
"Make the lighting like a studio"
"Point the camera at the scene, and make it isometric"

You can also hand over a reference image and have Claude reconstruct the equivalent Blender scene.


Persistent Credentials

Store API keys in Blender Add-on Preferences (Edit → Preferences → Add-ons → Blender MCP) — they survive Blender restarts. For headless/CI setups, inject via environment variables: BLENDERMCP_SKETCHFAB_API_KEY, BLENDERMCP_HYPER3D_API_KEY, BLENDERMCP_HUNYUAN3D_SECRET_ID, etc.


Security and Telemetry

Security: execute_blender_code runs arbitrary Python in Blender. Save your work before using it.

Telemetry: anonymous usage data collected by default. Opt out via:

"env": { "DISABLE_TELEMETRY": "true" }

Or uncheck in Blender’s Add-on preferences. Per the terms, data may be used to improve BlenderMCP, for research, and to train AI models.

Single instance: run only one MCP server at a time. The first command sometimes fails; subsequent ones work normally — expected behavior.


Why This Matters

Blender has a complete Python API. MCP exposes it to any LLM without touching Blender internals and without per-LLM adapters. 26K stars and an active ecosystem (scene reconstruction from screenshots, batch game asset generation via Hyper3D, headless 3D render pipelines) confirm the demand.

The pattern generalizes: any software with a programmable API can become an LLM-driven tool via MCP, no first-party integration required.


Mycelium Protocol — tracking the deep evolution of AI systems

© 2026 Mycelium Protocol. All rights reserved.

💬 评论与讨论

使用 GitHub 账号登录后发表评论

关于本站 · 免责声明

🍄 Mushroom Research Blog 是非营利、免费公开的个人科技观察博客与公众号 XStack18,不接受商业合作、不代表任何企业或机构立场,也不谋求商业利益。我们以个人视角客观中立地记录和分析 AI、Web3 等领域的最新模型发布与技术动态——不止转述新闻标题或二手信息,而是给出有独立思考的深入分析,希望帮更多人获得有价值的一手科技认知。

⚠️ 文中介绍的开源代码与模型,仅供学习交流与技术借鉴。它们大多仍处于早期阶段,有待进一步研究和验证,请勿直接用于工作或生产环境;如需采用,请先自行充分测试,并核实其许可证与安全性。
Open-source code and models featured here are shared for learning and reference only. Most are early-stage and still need further study and verification — please don't use them directly in your work or in production. Test them thoroughly and check their licenses and security first.

  1. 本站文章均为作者基于公开信息的个人研究与观点整理,不代表文中提及的任何公司、产品、模型的官方立场,未与其构成商业关联或合作关系。
  2. 科技行业信息更新极快,我们尽力保证内容准确、及时,但不对完整性、实时性做绝对保证,具体请以相关企业/项目官方公告为准。
  3. 文中引用的第三方商标、产品名称、图片、数据等版权归原权利人所有,我们会尽量注明来源;如你认为存在版权疑问或侵权,请通过下方邮箱联系我们,收到通知后会尽快核实处理(更正、加注来源或删除)。
  4. 文章内容仅为技术科普与个人观点,不构成投资、法律或其他专业建议,据此进行任何决策的后果需自行判断和承担。

📮 侵权 / 勘误 / 合作咨询:hello@mushroom.cv