Visiongemma

io.github.q1023884985Published by GitHub account q1023884985v2.1.0Updated Oct 24, 2026

Local AI vision MCP server for Windows: OCR, chart analysis, scene description. 100% offline.

VerifiedSTDIODesktop onlyMedia & DesignAI & ML

Overview

AI-generated overview

A local Windows vision MCP server that gives an AI assistant offline image analysis: OCR, chart reading and scene description via one vision_analyze tool.

What you can do
Analyze images: a single vision_analyze tool lets the assistant inspect pictures it cannot see itself, returning structured descriptions of product photos, screenshots and charts. Extract text: OCR of Chinese and English text from images. Interpret charts and describe scenes, all processed on the local machine. The same program also bundles a desktop pet that can chat and recognize the screen or images.
When to use it
Useful when the assistant receives images but has no image input of its own, and the work must stay on a Windows PC without sending pictures to a cloud service.
Requirements
Windows 10/11 x64 only; runs as a local stdio process launched from the VisionGemma.exe path with the --mcp argument. A Vulkan-capable GPU is recommended, with CPU fallback. About 5GB free disk space, since the roughly 3.2GB model is downloaded on first run and stored under %LOCALAPPDATA%.
Before you install
The project is marked deprecated and archived, with new users told not to use it. The README lists personal contact details (QQ, WeChat, email) and links to third-party sites; verify the download and its SHA256 before running. The desktop pet can recognize the screen, so be aware of what is visible on screen.

Installation

In SourceWeft

  1. Open Visiongemma in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.

Other MCP clients

Follow the launch instructions in the repository.

README

VisionGemma - 本地图片识别 + AI 桌宠

[!WARNING] 本项目已淘汰归档，不再维护。保留仓库与 Release 附件仅供已有用户追溯，新用户请勿使用。

在完全本地的 Windows 电脑上运行的开箱即用图片识别工具 + 趣味 AI 桌宠。图片识别全程在本机完成，不上传任何数据。

产品展示

Agent 自动调起识别

当你的 AI 助手遇到图片、又恰好不支持图片输入时，VisionGemma 会作为 MCP 工具被自动调起，把"看图能力"无缝交给它：

[Agent 自动调起识别]

一键接入你的 AI 助手

不用手写配置文件。软件内置对接引导，复制一段话粘贴给 AI 助手，或直接复制配置 JSON，即可完成 MCP 对接：

[一键 MCP 对接]

本地精准识别

选图 → 识别 → 得到结构化描述，全程离线。产品图、截图、图表都能看懂：

[本地精准识别]

趣味 AI 桌宠

桌面陪伴小宠物，支持自然对话、识别屏幕与图片：

[趣味 AI 桌宠]

功能特性

  • 傻瓜式部署：下载解压，双击运行，自动下载模型并完成配置，零命令行、零环境配置
  • Agent MCP 对接：可作为 MCP 服务器接入任意 AI 工具（vision_analyze 工具），让 AI 获得看图能力
  • 本地图片识别：图片分析、OCR 文字提取（中英文）、图表解读、场景描述全部在本机完成
  • Vulkan 加速：支持 NVIDIA / AMD / Intel 显卡硬件加速；无独显时自动回退 CPU 运行
  • 趣味 AI 桌宠：托盘启动可爱桌宠，可对话、可识别屏幕/图片，互动陪伴
  • 隐私安全：所有识别完全离线本地完成，不上传任何数据到云端
  • 单文件免安装：一个 exe 即完整运行，内置 Vulkan 引擎与桌宠

系统要求

  • Windows 10/11 x64
  • 显卡支持 Vulkan（NVIDIA / AMD / Intel 均可，推荐硬件加速）
  • 无独显时自动回退 CPU 运行
  • 磁盘剩余空间 >= 5GB（模型约 3.2GB）

下载

GitHub Release（推荐）

前往 Releases 页面下载 v2.1.1

安装包：VisionGemma-20260821.zip SHA256：83C6170BBD7367B1A535A586BEE08F6BAC73F5EAC089736E189B62017532B944

国内镜像

快速开始

  1. 解压压缩包，双击 VisionGemma.exe，首次运行自动打开配置向导
  2. 向导引导自动下载模型（约 3.2GB，支持断点续传）
  3. 下载完成后即可开始使用

趣味桌宠

  1. 右键托盘图标 →「趣味桌宠」，可快速启动桌宠，自动拉起本地识别引擎
  2. 桌宠支持对话、选择主题等功能
  3. 可通过托盘菜单设置开机自启

作为 MCP 服务器使用（接入 AI 工具）

在 AI 工具的 MCP 配置中加入：

json
{  "mcpServers": {    "vision-gemma": {      "command": "VisionGemma.exe 完整路径",      "args": ["--mcp"],      "type": "stdio"    }  }}

然后即可调用 vision_analyze 工具识别图片。更省事的做法：打开软件的「对接引导」，复制话术粘贴给 AI 助手即可。

文件说明

  • VisionGemma.exe 主程序（内置 Vulkan 引擎 + 桌宠，单文件免安装）
  • 模型存放于 %LOCALAPPDATA%\VisionGemma\models

隐私说明

所有图片识别完全在本地完成，不上传任何数据。

售后支持

友情链接

Source: README.md at commit 50a9188

Tools

0
SourceWeft has not started this package in its sandbox yet, so its tools are not known.

Version history

1
  1. v2.1.0LatestSep 16, 2026

Score

40% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 days25 pts

    Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.

  • Auth declared10 pts

    Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.

  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.