MCP Vision Server
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., "@MCP Vision Server describe what's in /home/user/photo.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.
MCP Vision Server · 视觉识别服务
基于 Kimi/Moonshot 视觉 API 的 MCP 服务器,作为 Claude Code 全局插件使用。传入本地图片路径,返回 AI 对图片内容的详细描述、文字提取等。
MCP server for image recognition via Kimi/Moonshot vision API. Works as a global Claude Code plugin.
中文
功能
describe_image — 识别图片内容,返回文字描述
describe_image_to_file — 识别并保存为 UTF-8 文件(解决 Windows 终端中文乱码)
支持 PNG / JPG / GIF / WebP / BMP,最大 20MB
支持自定义提示词(如"提取所有文字""描述图表结构")
安装
pip install mcp-vision-server或从源码安装:
git clone https://github.com/coffe-d/MCP-Vision-Server.git
cd mcp-vision-server
pip install -e .获取 API Key
在 Moonshot 开放平台 注册并创建 API Key。
注册到 Claude Code
claude mcp add vision-server \
--env KIMI_API_KEY="sk-你的密钥" \
-- mcp-vision-server注册后 Claude Code 即可使用 describe_image 和 describe_image_to_file 两个工具。
配置
环境变量 | 必填 | 默认值 | 说明 |
| 是 | — | Moonshot API 密钥 |
| 否 |
| API 地址 |
| 否 |
| 模型名称 |
工具说明
describe_image — 识别图片,返回文本描述。
参数 | 类型 | 必填 | 默认值 | 说明 |
| string | 是 | — | 图片绝对路径 |
| string | 否 | — | 自定义提示词 |
| int | 否 | 4096 | 最大输出长度 |
describe_image_to_file — 识别图片,结果保存为 UTF-8 文件。适合中文环境避免终端乱码。
参数 | 类型 | 必填 | 默认值 | 说明 |
| string | 是 | — | 图片绝对路径 |
| string | 否 | 自动(同名 .md) | 输出文件路径 |
常见问题
"KIMI_API_KEY environment variable is not set"
未设置环境变量。注册时确保使用了 --env KIMI_API_KEY="sk-..."。
终端中文乱码
使用 describe_image_to_file 代替 describe_image,结果直接写入 UTF-8 文件。
"不支持的图片格式"
仅支持 PNG、JPG、JPEG、GIF、WebP、BMP 格式。
许可
MIT — 详见 LICENSE。
Related MCP server: glm-vision-mcp-server
English
Features
describe_image — Recognize image content and return text description
describe_image_to_file — Recognize and save result to a UTF-8 file
Supports PNG / JPG / GIF / WebP / BMP up to 20MB
Customizable prompt for targeted extraction
Install
pip install mcp-vision-serverOr from source:
git clone https://github.com/coffe-d/MCP-Vision-Server.git
cd mcp-vision-server
pip install -e .Get an API key
Sign up at Moonshot Platform and create an API key.
Register with Claude Code
claude mcp add vision-server \
--env KIMI_API_KEY="sk-your-key-here" \
-- mcp-vision-serverConfiguration
Variable | Required | Default | Description |
| Yes | — | Moonshot API key |
| No |
| API base URL |
| No |
| Model name |
API Reference
describe_image — Return image description as text.
Parameter | Type | Required | Default | Description |
| string | Yes | — | Absolute path to image |
| string | No | — | Custom prompt |
| int | No | 4096 | Max output tokens |
describe_image_to_file — Save result to a UTF-8 file.
Parameter | Type | Required | Default | Description |
| string | Yes | — | Absolute path to image |
| string | No | auto (.md) | Output file path |
Troubleshooting
"KIMI_API_KEY environment variable is not set" — Make sure you passed --env KIMI_API_KEY="sk-..." when running claude mcp add.
Garbled Chinese in terminal — Use describe_image_to_file to write directly to UTF-8 file.
License
MIT — see LICENSE.
Available Tools
2 toolsdescribe_imageA
识别图片内容。传入本地图片的绝对路径,返回AI对图片内容的详细描述。
参数:
image_path: 图片的绝对路径,支持 PNG/JPG/JPEG/GIF/WEBP/BMP
prompt: 自定义提示词,可指定需要提取的信息类型(如"提取所有文字"、"描述图表结构"等)
max_tokens: 最大输出长度,默认4096
Inputs: image_path (required), prompt, max_tokens. Returns: result
describe_image_to_fileA
识别图片内容并保存到文件(解决 Windows 终端中文乱码问题)。
将AI识别结果直接写入UTF-8编码的文本文件,完全绕过终端编码问题。 如果未指定输出路径,默认在原图同目录下生成同名的 .md 文件。
参数:
image_path: 图片的绝对路径
output_path: 输出文件的路径(可选,默认与原图同名 .md)
prompt: 自定义提示词
max_tokens: 最大输出长度
Inputs: image_path (required), output_path, prompt, max_tokens. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 2 tools
Both tools describe images, but one outputs to terminal and the other saves to file. Their purposes are clearly distinguished, though the core functionality overlaps.
Both tools follow a consistent verb_noun pattern: 'describe_image' and 'describe_image_to_file'. The naming is predictable and clear.
With only two tools, the server feels thin but covers the basic need of describing images and optionally saving results. It is borderline for the typical 3-15 tool range.
The tool surface is minimal and lacks features like batch processing, model selection, or other vision tasks. Agents have no way to adjust output formats beyond file writing, leaving notable gaps.
Maintenance
Related MCP Connectors
Generate images and video from Claude with one API key.
Image & PDF tools for AI agents: compress, convert, resize, PDF, AI vision, pipeline.
Generate and edit images and create short videos inside Claude. Prepaid credits, no subscription.
OCR, transcription, file extraction, and image generation for AI agents via MCP.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables image analysis and understanding using Vision Language Models through OpenAI-compatible APIs. Supports analyzing images from URLs or local files with custom prompts.12MIT
- AlicenseAqualityDmaintenanceEnables Claude Code to analyze images using Zhipu AI's GLM-5V-Turbo vision model, supporting local files and URLs with customizable prompts.163 npmMIT
- FlicenseNot gradedqualityDmaintenanceEnables Claude Code to analyze images using multiple visual models (Kimi, OpenAI, etc.) and summarize results with DeepSeek. Supports dynamic switching of visual providers and handles both text and image inputs.1-
- FlicenseNot gradedqualityDmaintenanceEnables Claude Code to analyze images using Qwen vision models (via DashScope) when the main model is text-only.-