# Ultralytics Platform MCP

Connects Claude Code to Ultralytics Platform to manage datasets, train vision models, and fetch weights. Built for developers orchestrating ML runs from the chat.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · ★ 10
- Author: amanharshx
- License: MIT · policy=link-only · description only; the file stays with the author
- Source: https://github.com/amanharshx/ultralytics-mcp
- Updated: 2026-10-22
- HTML: https://affpapa.org/en/nevibecode/mcp/ultralytics-platform-mcp

## What it does

This server exposes Ultralytics Platform controls to your MCP client. It enables creating private workspaces, transferring local video files as datasets, launching fine-tuning runs for models like `yolo11n`, and checking epoch metrics. You can also trigger predictions on remote image URLs or save resulting weight files directly to your machine.

Running the package demands Node.js 20 or newer alongside a valid `ULTRALYTICS_API_KEY`. Uploading training material from local video clips depends on system binaries: `ffmpeg` and `ffprobe` must exist in your `PATH`. Without these media tools, video dataset ingestion fails immediately.

Jobs on the platform incur immediate charges. The server protects against unintended spend by enforcing the `confirm_cost: true` flag on training and export tools. Retrying training on an existing model checkpoint also requires `confirm_history_loss: true`, preventing accidental loss of recorded epoch logs.

## Requirements
- Needs: Node.js (commands run via npx)
- Environment variables: `ULTRALYTICS_API_KEY`, `ULTRALYTICS_API_BASE` — keys and tokens: ask the human for the values, never commit them or put them in shared files.

## Install

### Claude Code (native platform)

**Platform CLI**
```bash
claude mcp add --env ULTRALYTICS_API_KEY=<ULTRALYTICS_API_KEY> --env ULTRALYTICS_API_BASE=<ULTRALYTICS_API_BASE> --transport stdio ultralytics-platform-mcp -- npx -y ultralytics-mcp
```
Replace <…> with your own values

### Codex

**Platform CLI**
```bash
codex mcp add ultralytics-platform-mcp --env ULTRALYTICS_API_KEY=<ULTRALYTICS_API_KEY> --env ULTRALYTICS_API_BASE=<ULTRALYTICS_API_BASE> -- npx -y ultralytics-mcp
```
Replace <…> with your own values

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "ultralytics-platform-mcp": {
            "command": "npx",
            "args": [
                "-y",
                "ultralytics-mcp"
            ],
            "env": {
                "ULTRALYTICS_API_KEY": "<ULTRALYTICS_API_KEY>",
                "ULTRALYTICS_API_BASE": "<ULTRALYTICS_API_BASE>"
            }
        }
    }
}
```
Replace <…> with your own values

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "ultralytics-platform-mcp": {
            "type": "stdio",
            "command": "npx",
            "args": [
                "-y",
                "ultralytics-mcp"
            ],
            "env": {
                "ULTRALYTICS_API_KEY": "<ULTRALYTICS_API_KEY>",
                "ULTRALYTICS_API_BASE": "<ULTRALYTICS_API_BASE>"
            }
        }
    }
}
```
Replace <…> with your own values

### Gemini CLI

**Platform CLI**
```bash
gemini mcp add -e ULTRALYTICS_API_KEY=<ULTRALYTICS_API_KEY> -e ULTRALYTICS_API_BASE=<ULTRALYTICS_API_BASE> ultralytics-platform-mcp npx -y ultralytics-mcp
```
Replace <…> with your own values

## When to use it
- manage Ultralytics projects and datasets directly from chat
- upload local video files into platform datasets
- start and monitor fine-tuning runs for vision models
- download trained model weights to the local file system
- test model predictions using remote image links

## Example prompts
- Check my Ultralytics projects and tell me which datasets are ready for training.
- Create a new private project named `traffic-cams` and upload `./clips/junction.mp4` as data.
- Start a 50-epoch training run for `yolo11n` on the `traffic-cams` dataset.

## Similar
- [ml-model-training](https://affpapa.org/en/nevibecode/skills/secondsky-ml-model-training.md) — skill · 457 installs
- [huggingface-datasets](https://affpapa.org/en/nevibecode/skills/huggingface-datasets.md) — skill · 2400 installs
- [nemotron-customize](https://affpapa.org/en/nevibecode/skills/nemotron-customize.md) — skill · 2100 installs
- [data-researcher](https://affpapa.org/en/nevibecode/agents/data-researcher.md) — subagent · 542 installs
- [fiftyone-mcp-server](https://affpapa.org/en/nevibecode/mcp/fiftyone-mcp-server.md) — MCP server · ★ 26
- [firecrawl-knowledge-base](https://affpapa.org/en/nevibecode/skills/firecrawl-knowledge-base.md) — skill · 32600 installs

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md