Depth Anything MCP
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Depth Anything MCPestimate the depth map for ~/photos/street.jpg"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Depth Anything MCP
把 Depth Anything V2 做成 Cursor / Claude 可用的 MCP 工具。默认用 单文件 ONNX(约 95MB),不依赖 PyTorch。
用 uvx 一键接入(推荐,国内加速)
本机先装 uv(Windows:powershell -c "irm https://astral.sh/uv/install.ps1 | iex")。
国内直连 GitHub / PyPI 容易超时。把下面整段加进 Cursor 的 ~/.cursor/mcp.json,或项目里的 .cursor/mcp.json,然后 Reload MCP:
{
"mcpServers": {
"depth-anything": {
"command": "uvx",
"args": [
"--from",
"git+https://ghfast.top/https://github.com/HG-ha/depth-anything-mcp",
"depth-anything-mcp"
],
"env": {
"DEPTH_ANYTHING_HOME": "~/.depth-anything-mcp",
"DEPTH_ANYTHING_DEFAULT_ENCODER": "vits",
"DEPTH_ANYTHING_DEVICE": "auto",
"DEPTH_ANYTHING_MIRROR": "cn",
"HF_ENDPOINT": "https://hf-mirror.com",
"UV_DEFAULT_INDEX": "https://pypi.tuna.tsinghua.edu.cn/simple"
}
}
}
}这会走 GitHub 代理(ghfast)、Hugging Face 镜像(hf-mirror)和清华 PyPI。默认权重已打进包里,不用再单独下。代理失效可把 ghfast.top 换成 gh-proxy.com。海外用户把 --from 改回 git+https://github.com/HG-ha/depth-anything-mcp,并设 DEPTH_ANYTHING_MIRROR=off。
只要量化版或米制模型时才会额外下载:
$env:UV_DEFAULT_INDEX="https://pypi.tuna.tsinghua.edu.cn/simple"
uvx --from git+https://ghfast.top/https://github.com/HG-ha/depth-anything-mcp depth-anything-setup --variant quantizedRelated MCP server: OpenCV MCP Server
其它安装方式
PowerShell:
irm https://ghfast.top/https://raw.githubusercontent.com/HG-ha/depth-anything-mcp/main/install.ps1 | iexmacOS / Linux:
curl -fsSL https://ghfast.top/https://raw.githubusercontent.com/HG-ha/depth-anything-mcp/main/install.sh | bash或 pip:
pip install -i https://pypi.tuna.tsinghua.edu.cn/simple git+https://ghfast.top/https://github.com/HG-ha/depth-anything-mcp.git
python -m depth_anything_mcp.install本地仓库:.\install.ps1 / ./install.sh。
MCP 工具
工具 | 作用 |
| 图片深度(默认 ONNX) |
| 视频深度(默认逐帧 ONNX) |
| 模型清单 |
| 预下载 ONNX |
| 下载默认权重 |
| Runtime 状态 |
图片:
image_path指向本地文件或 URL默认白色深度图;彩色 MAGMA:
grayscale=false更小包:
variant=quantized室内米制:
metric_scene=indoor视频冒烟:
max_len=32
官方 Video Depth Anything 没有可打包的正式 ONNX,视频默认按帧跑 DA-V2。
模型
文件 | 大小 | 来源 |
| 94.5 MB | |
| 38.6 MB |
Small 为 Apache-2.0。Base / Large 为 CC-BY-NC-4.0。
GPU
DEPTH_ANYTHING_DEVICE=auto(默认)会在第一次启动时自动换 Runtime,用户不用自己对 CUDA 版本:
Windows:装
onnxruntime-directml,NVIDIA / AMD / Intel 都能走 GPU,不看 CUDALinux + NVIDIA:装 CUDA 12 档的
onnxruntime-gpu[cuda,cudnn](ORT>=1.21,<1.27,避免 1.27+ 默认 CUDA 13)失败或没有独显:回退 CPU
也可以在工具参数或环境变量里写 cpu / dml / cuda / cuda:1。强制 CPU:
"DEPTH_ANYTHING_DEVICE": "cpu"第一次走 GPU 时会换掉 CPU 版 onnxruntime 并拉对应 wheel,可能要一两分钟。国内默认走清华 PyPI。失败会回退 CPU,日志在 stderr。
开发
pip install -e ".[dev]"
python -m pytest许可
本仓库代码 Apache-2.0。模型权重以官方卡片为准。
This server cannot be deployed
Maintenance
Related MCP Connectors
Video, audio, and image processing for AI agents: convert, transcribe, upscale - 150+ operations.
FFmpeg as a service for AI agents: typed video editing tools, async jobs, downloadable outputs.
CVE lookups (NVD) and dependency-manifest audits (OSV) for AI agents. No API keys.
CVE lookups (NVD) and dependency-manifest audits (OSV) for AI agents. No API keys.
Related MCP Servers
- AlicenseNot gradedqualityCmaintenanceProvides camera and vision tools for AI assistants to list available cameras, capture images from USB cameras, and save frames to disk for use with LLMs.4MIT
- AlicenseNot gradedqualityFmaintenanceProvides OpenCV computer vision capabilities for AI assistants, enabling image processing, object detection, face detection, and video analysis through natural language.111MIT
- FlicenseNot gradedqualityDmaintenanceEnables AI assistants to perform monocular depth estimation on images using Apple's Depth Pro model, with tools for single or batch processing and GPU management.2-
- AlicenseAqualityAmaintenanceEnables text-only AI coding agents to analyze images and videos via vision-capable models (Gemini, Grok, OpenRouter), returning text descriptions for reasoning.213MIT