label-studio

This MCP server provides programmatic access to a Label Studio instance via the label-studio-sdk. It exposes tools to manage projects (create, list, view details and XML labeling configuration), manage tasks (import tasks, list tasks, fetch task data and annotations), and add model predictions to tasks. You can issue natural language or structured calls through MCP clients to perform common Label Studio operations, such as creating a new labeling project, importing a set of tasks from a JSON file, retrieving task payloads, or attaching model predictions to specific tasks. The server relies on the official Label Studio SDK to communicate with your Label Studio instance, so it requires a reachable Label Studio URL and a valid API key.

To use the server, ensure you have a running Label Studio instance and the MCP server configured with your Label Studio API key and URL. The provided tools (get_label_studio_projects_tool, create_label_studio_project_tool, import_label_studio_project_tasks_tool, create_label_studio_prediction_tool, etc.) can be invoked by MCP clients to perform the listed operations. For example, you can create a new project by calling create_label_studio_project_tool with a title and XML labeling configuration, import tasks via import_label_studio_project_tasks_tool from a JSON file, and then attach predictions to specific tasks using create_label_studio_prediction_tool.

Prerequisites:\n- A running Label Studio instance accessible from the MCP server host.\n- A Label Studio API key with appropriate permissions.\n- Python environment with uv (uvx) tooling available.\n\nInstallation steps:\n1) Clone the MCP server repository:\n git clone https://github.com/HumanSignal/label-studio-mcp-server.git\n cd label-studio-mcp-server\n\n2) Set up a Python virtual environment and install dependencies (example using uv/uvx workflow shown in the README):\n # Create and activate a virtual environment (adjust for your OS)\n uv venv\n source .venv/bin/activate # on Unix-like systems\n \n # Install and synchronize dependencies (if applicable)\n uv sync\n\n3) Ensure you have a Label Studio instance running and obtain an API key.\n - LABEL_STUDIO_API_KEY: your_api_key_here\n - LABEL_STUDIO_URL: http://localhost:8080 (or your Label Studio URL)\n\n4) Run the MCP server configuration (example as shown in the README):\n The following configuration can be added to your client configuration file to start the MCP server via uvx: \n {\n "mcpServers": {\n "label-studio": {\n "command": "uvx",\n "args": [\n "--from",\n "git+https://github.com/HumanSignal/label-studio-mcp-server",\n "mcp-label-studio"\n ],\n "env": {\n "LABEL_STUDIO_API_KEY": "your_actual_api_key_here",\n "LABEL_STUDIO_URL": "http://localhost:8080\"\n }\n }\n }\n }\n\n5) Start the MCP server using your environment’s standard startup process (as configured by your MCP client).

Tips and caveats:\n- Ensure LABEL_STUDIO_URL is reachable from the MCP server host, and that the API key has sufficient permissions for the requested operations.\n- If you rotate API keys, update the environment variables accordingly.\n- The label-studio-sdk-based tools expect specific parameters (e.g., project_id, task_id) when calling tools like get_label_studio_project_details_tool or create_label_studio_prediction_tool; consult the Tools section in the README for exact signatures.\n- When importing tasks from a JSON file, the file should contain a list of task objects compatible with Label Studio.\n- If your Label Studio instance is secured behind a firewall or requires a proxy, ensure network rules allow the MCP server to reach LABEL_STUDIO_URL.\n- It’s common to run Label Studio on a separate port; adjust LABEL_STUDIO_URL accordingly and ensure CORS or API access settings permit MCP server access.