MCP YOLOE: Zero-Shot Object Detection & Segmentation
@rjn32s
Provide your AI agents with "eyes." This server enables open-vocabulary object detection and instance segmentation using naturally phrased text prompts (e.g., "detect the laptop next to the coffee").
Overview
What is MCP YOLOE: Zero-Shot Object Detection & Segmentation?
MCP YOLOE is a Model Context Protocol server that gives AI agents advanced computer vision using zero-shot learning. It detects and segments any object described in natural language, unlike fixed-list YOLO models.
How to use MCP YOLOE: Zero-Shot Object Detection & Segmentation?
You interact through an AI agent (e.g., Claude) by providing a natural language prompt such as “Find the ‘vintage typewriter’ in this image and give me its exact coordinates.” The server accepts local file paths, remote image URLs, and Base64-encoded images as input.
Key features of MCP YOLOE: Zero-Shot Object Detection & Segmentation
- Zero-shot detection with natural language prompts.
- Precision segmentation with exact polygon masks.
- Flexible inputs: local paths, URLs, Base64.
- Agent-first design for Claude, IDEs, and workspace agents.
- Uses YOLOE26‑L architecture (55.0 mAP, ~6.2ms on T4).
Use cases of MCP YOLOE: Zero-Shot Object Detection & Segmentation
- Find and locate arbitrary objects in images using descriptive text.
- Obtain exact polygon masks for detected objects for further processing.
- Integrate computer vision capabilities into AI agents for automated image analysis.
- Support diverse input sources (local files, URLs, embedded data) in agent workflows.
FAQ from MCP YOLOE: Zero-Shot Object Detection & Segmentation
How is this different from traditional YOLO models?
Traditional YOLO only detects a fixed list of objects, while this server uses zero-shot learning to detect anything described in natural language.
What is the underlying model and its performance?
It uses the state-of-the-art YOLOE26‑L architecture, achieving high precision (55.0 mAP) and rapid inference (~6.2ms on T4 GPUs).
What input formats can I use?
The server works with local file paths, remote image URLs, and Base64-encoded strings.
Is it designed for agent integration?
Yes, it is specifically built for integration with Claude, IDEs, and autonomous workspace agents.
Does it require special runtime or dependencies?
The README does not specify runtime requirements beyond the model architecture (YOLOE26‑L).
