io.github.q1023884985/visiongemma — Local AI vision MCP server
This MCP server provides local AI image recognition on Windows. It performs OCR, chart analysis, and scene description entirely on-device (100% offline), producing structured descriptions. It can be invoked by an AI agent when image input isn’t directly supported, and it supports an easy MCP connection flow via built-in guidance or copying configuration JSON.
🛠️ Key Features
- Local Windows image recognition (OCR, chart analysis, scene description)
- 100% offline operation; no data upload
- Agent MCP integration for automated “look at image” tool invocation
- Offline processing to return structured descriptions
- “Download, unzip, double-click” deployment with automatic model download and configuration
🚀 Use Cases
- OCR for screenshots and local images
- Understanding product images, charts, and diagrams
- Enabling an AI assistant to handle image inputs via MCP tooling
- Desktop use with an AI companion that can recognize screens and images
⚡ Developer Benefits
- MCP tool can be automatically triggered by an agent when image input is unsupported
- MCP setup via guided onboarding or by copying configuration JSON
- Structured output descriptions for downstream use
⚠️ Limitations
- Described deployment and operation target is Windows (local).
