Roboflow + Codex
Run inference, build Workflows and manage datasets and models in Roboflow from Codex, with the Roboflow plugin and MCP server.
How to connect Codex to Roboflow
5 stepsOpen plugins in Codex
In the Codex app, click the Plugins icon in the left pane. In the Codex CLI, run codex and enter /plugins.
Install the Roboflow plugin
Search for Roboflow and install it. In the CLI, press Space to enable it if it is installed but disabled.
Start a new session and sign in
Start a new Codex session. On the first Roboflow call, sign in with your Roboflow account and approve the permissions.
Check the connection
Send "@Roboflow List the projects in my workspace" and confirm Codex returns your projects.
Ask Codex to build
Ask Codex to run a model on your images, build a Workflow, or write the code to call your model from your app.
What teams build with Codex
Test a model on your own images
Run detection or segmentation on files in your repo and see the annotated results.
Add vision to an app
Have Codex write Python that calls your Roboflow model through the hosted API.
Build a Workflow from a description
Describe the pipeline to Codex, then publish the Workflow and copy its deployment code.
Questions
No. Codex signs in with your Roboflow account. For headless servers where a browser sign-in isn't possible, you can connect with an API key instead.
No. The MCP server and Skills have no extra cost. You pay only for the Roboflow usage behind them on your plan.
Yes. Any MCP client that supports streamable HTTP can connect to https://mcp.roboflow.com/mcp.
Why are companies choosing Roboflow?
The end-to-end platform for vision AI: create datasets, train models and deploy to production.
Flexible deployment modes
Move from a scalable cloud API to edge hardware like NVIDIA Jetson without managing complex dependencies.
Integrated workflow builder
Combine multiple AI models, custom business logic and third-party integrations into one production-ready pipeline in a low-code editor.
Model monitoring and alerts
Dashboards track confidence, inference time and request volume for your deployed models, and email alerts tell your team when confidence suddenly drops.
