Visionator
Empowers agentic models lacking native vision support to analyze images using local Ollama vision models via an MCP server.
Visionator is a lightweight Model Context Protocol (MCP) server built with C# and .NET, specifically designed to extend vision capabilities to AI agents like GitHub Copilot. It enables models without inherent vision or multimodal support to perform detailed image analysis by integrating with local Ollama vision models. This tool bypasses the need for Python or Node.js environments, offering a simple, self-contained solution. Its primary focus is on Unity scene and game screenshot analysis for QA, debugging, and UI verification, but it is versatile enough for any image analysis task.
Características Principales
01Flexible Image Input: Accepts images via file paths or base64-encoded strings, simplifying agent interaction with a helper tool for direct file reading.
02Intelligent Image Preprocessing: Automatically resizes, normalizes, and validates images using SkiaSharp, optimizing them for vision model inference and providing pixel-level comparison for accurate diffing.
03Comprehensive Vision Tools: Includes specialized tools for QA-focused analysis, free-form Q&A, area-specific descriptions, text extraction (OCR), object validation, screenshot comparison, and color palette analysis.
040 GitHub stars
05Ollama Integration & Management: Connects seamlessly with local Ollama instances, supporting flexible configuration via environment variables and offering an optional auto-start feature for a zero-config setup.
06Agent-Optimized Output: Strips common 'thinking' or 'chain-of-thought' tags from model responses to deliver concise, actionable outputs directly to the agent.
Casos de Uso
01Perform visual validation and regression testing for game scenes and application UIs.
02Debug rendering issues in game development, such as identifying missing textures or broken shaders.
03Conduct automated quality assurance (QA) and UI inspection of game and application screenshots.