How to integrate Roboflow Universe using MCP

Run hosted computer-vision model inference. This guide creates an MCP server from Arthur's Roboflow Universe template — tools preconfigured against https://serverless.roboflow.com — and finishes by sharing the server's MCP swagger documentation.

Steps

Open the Secrets vault and create a secret named ROBOFLOW_UNIVERSE_API_KEY holding your API key. Create a Roboflow workspace key and send it in the api_key query parameter. Keep private keys in Arthur MCP Secrets.
Go to the REST API Templates gallery and search for Roboflow Universe.
Click Use template on the Roboflow Universe card to open the server-creation dialog.
Review the server name and select ROBOFLOW_UNIVERSE_API_KEY in the credential field — the dialog only accepts vault secrets, and it lists every tool that will be created.
Click Create server. Arthur creates the server, applies the authentication, and generates all the tools.
Open the server's Tools tab to review the generated tools.
Open the server's Connect section and click Share the MCP swagger documentation — Arthur generates the public documentation link and QR code for this server.

Confirm it worked

The server appears with 1 preconfigured tool(s): run_object_detection. The Connect section offers the MCP swagger documentation — a shareable page with setup instructions for any AI client.

Good to know

Run a Roboflow object-detection model against a hosted image URL and return JSON predictions containing classes, confidence, and bounding boxes.
This integration requires a credential (api-key). Get one at https://app.roboflow.com/.
Official API documentation: https://docs.roboflow.com/deploy/serverless-hosted-api-v2/use-with-the-rest-api
Editing a tool later never changes the template — templates are starting points, and the server is fully yours after creation.

Tutorial video

How to integrate Roboflow Universe using MCP

Steps

Confirm it worked

Good to know

Related