Label Studio MCP Server

🚀 Label Studio MCP Server

This project provides a Model Context Protocol (MCP) server that allows interaction with Label Studio instances via the label-studio-sdk. It supports programmatic management of annotation projects, tasks, and predictions through natural language or structured MCP client calls.

🚀 Quick Start

The Label Studio MCP Server enables seamless interaction with Label Studio instances. With it, you can perform operations such as creating projects, managing tasks, and adding predictions using natural language or structured MCP client calls.

✨ Features

• Project Management: Create, read, update, and delete annotation projects.
• Task Management: Import task files, list all tasks, and retrieve data and annotations for specific tasks.
• Prediction Functionality: Generate and add prediction results for specific tasks.

📦 Installation

Requirements

1. Operating System: Linux or macOS is supported (Windows is not currently supported).
2. Python Version: 3.6 or higher.
3. Dependencies:
   • label-studio-sdk >= 0.9.0
   • uvicorn >= 0.15.0
   • python-multipart >= 0.2.1

Steps

Install the required dependencies using the Python package manager:

pip install label-studio-sdk uvicorn python-multipart

📚 Documentation

Configuration Guide

Environment Variable Configuration

Before running the server, ensure that the following environment variables are set:

• LABEL_STUDIO_URL: The URL of the Label Studio service (e.g., http://localhost:8000).
• LABEL_STUDIO_API_KEY: The API key for accessing the Label Studio API.

Example Configuration

export LABEL_STUDIO_URL="http://localhost:8000"
export LABEL_STUDIO_API_KEY="your_api_key_here"

Server Startup

Start the MCP server using the following command:

python label-studio-mcp.py --transport http --port 3000

Parameter Explanation:

• --transport {http|stdio}: Specify the communication transport method (default is stdio).
• --port PORT: Set the port number for HTTP transport (default is 3000).
• --host HOST: Set the host address for HTTP transport (default is 0.0.0.0).

Tool Explanation

Project Management Tools

• create_label_studio_project: Create a new annotation project.
• get_label_studio_projects: Retrieve all existing annotation projects.
• update_label_studio_project: Update the configuration of a specified project.
• delete_label_studio_project: Delete a specified annotation project.

Task Management Tools

• import_label_studio_tasks: Import tasks from a JSON file into a specified project.
• list_label_studio_tasks: List all tasks in a specified project.
• get_label_studio_task_data: Retrieve data and metadata for a specific task.

Prediction Functionality Tools

• create_label_studio_prediction: Generate and add prediction results for a specific task.
• list_label_studio_predictions: List all prediction results for a specified task.

💻 Usage Examples

Basic Usage

1. Create a new project:

create_label_studio_project --name "Image Classification Task" --description "Image classification annotation"

2. Prepare a task file (e.g., tasks.json):

[
  {
    "id": 1,
    "data": {
      "image": "path/to/image.jpg",
      "text": "Describe the image content"
    },
    "annotations": []
  },
  {
    "id": 2,
    "data": {
      "image": "path/to/another_image.jpg",
      "text": "Description of another image"
    },
    "annotations": []
  }
]

3. Import tasks into a project:

import_label_studio_tasks --project-id 1 --tasks-file tasks.json

4. List all tasks:

list_label_studio_tasks --project-id 1

5. Retrieve data for a specific task:

get_label_studio_task_data --task-id 1

Advanced Usage

6. Generate prediction results:

[
  {
    "result": [
      {
        "id": 1,
        "start_time": 0.1,
        "end_time": 0.2,
        "label": "cat"
      },
      {
        "id": 2,
        "start_time": 0.3,
        "end_time": 0.5,
        "label": "dog"
      }
    ],
    "score": 0.89
  }
]

7. Add predictions:

create_label_studio_prediction --task-id 1 --result-file predictions.json

📄 Contact

If you need help or have feedback, please submit an issue via GitHub Issues.