MCP-YOLO
MCP-YOLO is an agent-first development platform that provides Zero-Shot Object Detection and Segmentation as a Model Context Protocol (MCP) server. Powered by Ultralytics YOLOE, it enables developers and AI agents to detect and segment objects using arbitrary text prompts without retraining.
Key Features
- Zero-Shot Detection: Detect any object using natural language (e.g., "the blue coffee cup next to the spoon").
- Instance Segmentation: Precise polygon masks for discovered objects.
- Flexible Image Inputs: Supports local file paths, remote URLs, and Base64 encoded strings.
- Agent Optimized: Includes custom "Skills" for autonomous deployment and benchmarking.
YOLOE Performance Reference
YOLOE builds upon the latest YOLO architectures (like YOLO11 and YOLO26) to provide state-of-the-art open-vocabulary performance.
| Model | Based On | mAP (COCO) | Speed (T4/ms) | Params (M) |
|---|---|---|---|---|
| YOLOE26-N | YOLO26-N | 40.9 | 1.7 | ~3.0 |
| YOLOE26-S | YOLO26-S | 48.6 | 2.5 | ~10.0 |
| YOLOE26-L | YOLO26-L | 55.0 | 6.2 | ~40.0 |
| YOLOE-L | YOLO11-L | ~52.0 | ~5.0 | ~26.0 |
Note: Performance varies depending on the hardware and input resolution. mcp-yolo uses yoloe-26l-seg.pt by default for high precision.
Quick Start
Installation
Running the Server
MCP Tools
detect_objects
Performs zero-shot detection.
- Arguments:
image_source(str): Path, URL, or Base64.classes(list[str], optional): Custom text prompts to detect.
segment_objects
Performs zero-shot instance segmentation.
- Arguments:
image_source(str): Path, URL, or Base64.classes(list[str], optional): Custom text prompts to segment.
Publishing
This project is configured for automated PyPI publishing. See the pypi_setup_guide.md for details.