home / mcp / fiftyone mcp server
Expose FiftyOne datasets and operators to AI assistants via MCP for natural language workflows.
Configuration
View docs{
"mcpServers": {
"adonaivera-fiftyone-mcp-server": {
"command": "fiftyone-mcp",
"args": []
}
}
}You can run the FiftyOne MCP Server to expose FiftyOne’s dataset and operator capabilities to AI assistants. This lets you manage datasets, run operators, install plugins, and orchestrate workflows through natural language conversations. The server supports multiple client configurations and can be used with Claude, Cursor, VSCode, and ChatGPT Desktop environments.
After you configure a client to connect to the FiftyOne MCP Server, restart your AI assistant. You can then ask it to list datasets, load a dataset and show its summary, explore available operators for dataset management, set the current dataset context, install plugins, or find similar images within your dataset. The system is designed for multi-step, natural language workflows, so you can chain commands like loading a dataset, applying an operator, and then saving results in a single conversation.
Key usage patterns include: listing all datasets, loading a specific dataset and requesting its summary, querying which plugins are available and installing one, and initiating a search for similar images within your dataset. You can also delegate execution to a FiftyOne App session for actions that require a GUI-based workflow.
Prerequisites: you need Python 3.8+ and a working Python environment with pip. You also should have an AI assistant client capable of MCP configuration (Claude Desktop, Claude Code, Cursor, VSCode, or ChatGPT Desktop).
Step 1: Install the FiftyOne MCP Server package using pip.
pip install fiftyone-mcp-serverStep 2: Configure your MCP client to connect to the server. Choose one of the following configuration blocks that matches your environment.
{
"mcpServers": {
"fiftyone": {
"command": "fiftyone-mcp"
}
}
}If you have uv installed, you can run the MCP server directly without a separate installation step.
{
"mcpServers": {
"fiftyone": {
"command": "uvx",
"args": ["fiftyone-mcp-server"]
}
}
}List all available FiftyOne datasets in the current workspace.
Load a specific dataset by name or identifier and make it the active context.
Generate a summary of the loaded dataset, including counts, labels, and views.
Dynamically run any FiftyOne operator against the active dataset and view.
Query and discover FiftyOne plugins that enhance functionality.
Install a requested FiftyOne plugin to extend capabilities.
Open the FiftyOne App for delegated execution and GUI-driven tasks.
Control and manage sessions to delegate execution and share context across tools.