Blender MCP
Enables any MCP client to drive Blender 5.2 LTS through natural language, with tools for scene inspection, object creation and transformation, material and modifier handling, rendering, viewport capture, and guarded Python execution.
README
Blender MCP — Streamable-HTTP bridge for Blender 5.2 LTS
A Model Context Protocol bridge that lets any MCP client (Antigravity, Cursor, Claude Desktop, VS Code 1.101+, MiniMax Code, Windsurf, Continue, Cline, OpenCode, Codex, MCP Inspector, …) drive Blender 5.2 LTS through natural language.
The add-on runs an MCP Streamable-HTTP server inside Blender on a background thread. The client just points at it — no subprocess is spawned, no socket pair to babysit.
Why this shape
- Single install path inside Blender. The MCP server runs in-process
using
bpy, so there's nouvx/pipxdance and no socket pair to keep in sync. - Streamable HTTP on localhost. Works with every modern MCP client.
- Hand-rolled MCP server in the add-on. Blender 5.2 LTS bundles
Python 3.13 with no external packages by default. A ~300-line MCP
server is more portable than dragging in
mcp[cli]and editingsys.path. - Client-agnostic. No proprietary plugin format. Any MCP client that can hit an HTTP endpoint (or a stdio proxy that does) works.
Project layout
blender_addon_mcp/
├── README.md ← you are here
├── LICENSE ← MIT
├── pyproject.toml ← dev tooling for the add-on (ruff, mypy, pytest)
├── .gitignore
├── addon/
│ └── blender_mcp_addon/ ← zips into a Blender-installable .zip
│ ├── __init__.py ← bl_info, register/unregister, preferences
│ ├── server/
│ │ ├── __init__.py
│ │ ├── mcp_server.py ← MCP protocol (initialize, tools/list, tools/call, ping)
│ │ ├── transport.py ← stdlib http.server, runs on a daemon thread
│ │ └── protocol.py ← JSON-RPC 2.0 + MCP 2025-03-26 types
│ ├── tools/
│ │ ├── __init__.py ← tool registry
│ │ ├── scene.py ← scene_info, frame control
│ │ ├── objects.py ← list, get, create, delete, transform
│ │ ├── materials.py ← create, assign, modify
│ │ ├── modifiers.py ← add, set params, remove
│ │ ├── render.py ← render_image, screenshot_viewport
│ │ └── code.py ← execute_python (with confirmation)
│ └── utils/
│ ├── __init__.py
│ ├── bpy_helpers.py ← bpy → JSON serialization
│ └── safe_eval.py ← ast-validated code execution
├── scripts/
│ ├── install_addon.ps1 ← zip and install into Blender
│ ├── blender_mcp_proxy.js ← optional stdio/socket ↔ HTTP bridge for
│ │ Antigravity builds that don't accept `url`
│ └── start_blender_mcp_bridge.ps1 ← start/stop/status lifecycle for the bridge
├── examples/
│ ├── mcp-configs.md ← copy-paste configs for every major MCP client
│ ├── basic_scene.md
│ ├── material_setup.md
│ └── render_workflow.md
└── docs/
├── architecture.md
├── blender_5_2_notes.md
└── tools_reference.md
Quickstart
1. Install the add-on
# clone
git clone https://github.com/tattooinmtl/Blender_MCP.git
cd Blender_MCP
# install the add-on into every Blender version on this machine
pwsh ./scripts/install_addon.ps1
2. Enable the server in Blender
- Open Blender.
- Edit → Preferences → Add-ons → search "Blender MCP" → check the box.
- Set the port (default
8765) and click Start Server (or use the Blender MCP panel in the 3D Viewport sidebar). - The console should print:
[Blender MCP] Server listening on http://127.0.0.1:8765/mcp
3. Wire up your MCP client
Pick the client you use and paste the matching snippet from
examples/mcp-configs.md. The Antigravity
url form is the simplest:
{
"mcpServers": {
"blender": { "url": "http://127.0.0.1:8765/mcp" }
}
}
Snippets for Cursor, Claude Desktop, VS Code, MiniMax Code, Windsurf, Continue, Cline, and the MCP Inspector are all in that file.
4. Verify
In any MCP client chat:
Use the blender server: ping.
You should see the agent call tools/call on ping and report an empty
successful result. From a terminal:
curl -s -X POST http://127.0.0.1:8765/mcp `
-H "Content-Type: application/json" `
-d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"probe","version":"0"}}}'
Tools exposed
| Tool | Purpose | Read-only |
|---|---|---|
get_addon_info |
Capability handshake — name, version, protocol, tools list. Call this first. | yes |
scene_info |
Scene summary (objects, lights, cameras, frame) | yes |
list_objects |
List scene objects, optional type filter | yes |
get_object |
Full object details (location, rotation, scale, modifiers, materials) | yes |
create_object |
Create primitive (cube, sphere, plane, cylinder, cone, torus, empty, camera, light) | no |
delete_object |
Delete an object by name | no |
transform_object |
Set location/rotation/scale | no |
set_material |
Create or assign a material with a Principled BSDF | no |
set_texture |
Apply an image file to a material input (base_color, roughness, normal, etc.) | no |
import_asset |
Import GLB/GLTF/FBX/OBJ/STL/PLY/DAE/USD/X3D/ABC | no |
add_modifier |
Add a modifier (subdivision, mirror, solidify, array, bevel) | no |
remove_modifier |
Remove a modifier by name | no |
frame_get / frame_set |
Read / move the timeline | mixed |
render_image |
Render the current frame to a file | no |
screenshot_viewport |
Capture the current viewport | no |
execute_python |
Run arbitrary bpy code (guarded by ast validation) |
no |
ping |
Health check | yes |
See docs/tools_reference.md for the full schema
and examples.
Development
# syntax-check the add-on (Blender's Python is at <blender>/python/bin/python.exe)
python -m py_compile addon/blender_mcp_addon/__init__.py
python -m py_compile addon/blender_mcp_addon/server/mcp_server.py
# ...etc
# lint
ruff check addon/
mypy addon/
# test the MCP server in isolation (no bpy required)
pytest addon/tests/ -v
License
MIT. See LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。