Metadata-Version: 2.4
Name: visiongemma-mcp
Version: 2.1.0
Summary: Local AI vision MCP server for Windows: OCR (Chinese & English), chart analysis, scene description, object recognition. 100% offline — images never leave your machine.
Author: q1023884985
License: Proprietary
Keywords: mcp,vision,ocr,llm,computer-vision,local
Classifier: Development Status :: 4 - Beta
Classifier: Environment :: Win32 (MS Windows)
Classifier: Intended Audience :: Developers
Classifier: Operating System :: Microsoft :: Windows
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.9
Description-Content-Type: text/markdown

# VisionGemma MCP

mcp-name: io.github.q1023884985/visiongemma

**VisionGemma** is a local AI vision recognition MCP server for Windows. It exposes a single `vision_analyze` tool that lets any MCP client (Claude Desktop, Cursor, Cline, opencode, ...) do:

- **OCR** — extract text from images and screenshots (Chinese & English)
- **Chart / diagram analysis** — interpret tables, plots, diagrams and screenshots
- **Scene description** — describe what is in an image
- **Object recognition** — identify objects, people, expressions, atmosphere

All inference runs **100% locally** via a Vulkan-accelerated on-device engine. Images never leave your machine.

> **Note:** this package is the stdio launcher for the MCP Registry. It requires the VisionGemma engine (Windows x64) to be installed. Download it from the [GitHub Releases](https://github.com/q1023884985/visiongemma/releases), then run the installer once.

## Installation

```
pip install visiongemma-mcp
```

Install the VisionGemma engine from the [GitHub Releases](https://github.com/q1023884985/visiongemma/releases) page.

## MCP configuration

Add to your MCP client config:

```json
{
  "mcpServers": {
    "vision-gemma": {
      "command": "visiongemma",
      "args": ["--mcp"],
      "type": "stdio"
    }
  }
}
```

Or point directly at the engine executable (no pip install needed):

```json
{
  "mcpServers": {
    "vision-gemma": {
      "command": "C:\\path\\to\\VisionGemma.exe",
      "args": ["--mcp"],
      "type": "stdio"
    }
  }
}
```

## Tool: vision_analyze

| Parameter | Type | Description |
|---|---|---|
| `image_path` | string (required) | Local absolute path, `http(s)://` URL, or base64 data URI of the image |
| `prompt` | string (optional) | Question or instruction about the image (default: "please describe this image") |

## Privacy & licensing

- All processing is local and offline — no data leaves your device.
- Free 7-day trial; 38 CNY one-time purchase to keep it forever. See in-app activation.
- Windows 10/11 x64, Vulkan-capable GPU (or CPU fallback).
