May 21
Loading video player...
In this video, Tony França, Engineer at Roboflow, introduces the Roboflow MCP Server, a new way to connect your AI coding agent to Roboflow so you can build and deploy computer vision applications that would otherwise be out of reach. The MCP Server works with any agent that supports the Model Context Protocol, including Claude Code, Codex, and Cursor, and it exposes Roboflow's tools and skills directly inside your chat session so you can do almost everything you would normally do in the Roboflow UI from a single conversation. Tony walks through installation in under a minute. You grab your API key from your workspace settings, paste a single command into your terminal for Claude Code, or edit the config file for Codex. Then he jumps into a live demo. Starting from a local folder of solar panel images, Tony asks Claude Code to help him build a defect detection application, and the agent takes it from there. It creates a new object detection project, zips and uploads the images, kicks off auto-labeling, searches Roboflow Universe for related public datasets, forks a matching dataset into the workspace, generates a version, and starts a training run. = Additional Resources = Roboflow MCP Server: https://mcp.roboflow.com Getting Started with Roboflow: https://docs.roboflow.com/ = Chapters = 00:00 Introduction: Computer Vision Powers for Your Coding Agent 00:33 What is the Roboflow MCP Server? 01:26 Installation: Connecting Claude Code and Codex with Your API Key 02:45 Demo Setup: A Folder of Solar Panel Defect Images 03:19 Asking Claude to Build a Defect Detection App 05:07 Creating the Project and Uploading Images 06:15 Auto-labeling with the Sentry Foundation Model 07:26 Searching Roboflow Universe for a Better Dataset 09:55 Forking a Public Dataset into Your Workspace 11:30 Generating a Version and Training a Model 12:53 Composability: Combining MCP Servers Across Your Tools 13:30 A More Complete Example: Building a Web App with the Roboflow API 14:21 Tool Overview: Projects, Uploads, Auto-label, Training, and Evaluation 15:01 Get Started Today