• LABELS
  • COORDINATES
  • SCORES
  • HUMAN

How to test model predictions with your AI

Connect Clarifai to your agent to run live inference on images. You can check how your models identify objects and draw bounding boxes immediately.

Ask AI about this page

Short answer

How can I test my AI model's image predictions?

Give your agent an image URL to run a prediction through Clarifai. This returns specific labels and bounding box coordinates to help you check accuracy. You'll see the specific outcomes below.

Validation outcomes

What you get from inference

These are the specific data points your agent returns after a prediction.

LABELS

Classification tags

The model returns specific text tags for the image. You can see if the identified objects match what you expect.

COORDINATES

Bounding box data

The agent provides the spatial location of detected objects. This shows you exactly where the model thinks an object is located.

SCORES

Confidence levels

Each prediction includes a probability score. You can use these to judge how certain the model is about its findings.

HUMAN

Manual verification

If the scores are low or the boxes are off, you decide if the prediction is correct or needs adjustment.

The workflow

What your AI does when an image URL arrives.

Your agent handles the communication with Clarifai to process the visual data.

  1. Receive URL

    Your agent takes the image link you provide and prepares the request.

    predict_model
  2. Run inference

    The agent calls the model to analyze the image content.

    predict_model
  3. Extract data

    The agent pulls the classification tags and bounding box coordinates from the response.

    predict_model
  4. Report results

    Your agent presents the final labels and spatial data in your chat window.

    predict_model

Try it

Copy these to start.

Use these prompts to begin testing your model immediately.

Starting points

Replace the example URLs with the actual image links you want to test.

Clarifai Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_Wl7k3LILBATZT8oR2K1AY33RSlJewMKygOPyhcuJ/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Clarifai capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "clarifai-vision-ai-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_Wl7k3LILBATZT8oR2K1AY33RSlJewMKygOPyhcuJ/mcp"
    }
  }
}
Copy into chat04
  • Predict the contents of this image: https://example.com/sample-image.jpg

  • Check the bounding box accuracy for the objects in this URL: https://example.com/object-test.png

  • Run a classification on this image and tell me the confidence scores: https://example.com/test.jpg

  • Does this model detect a dog in this image? https://example.com/dog.jpg

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Clarifai once, then ask.

Link your account once and your credentials stay encrypted. You can then run predictions in any MCP-compatible client.

Connect Clarifai to your AI

FAQ

How this task behaves

  • 01

    Can the AI train new models?

    No. The agent can only run predictions on existing models.

  • 02

    Can I upload files directly from my computer?

    No. You must provide a URL that the agent can access.

  • 03

    Can the AI change my Clarifai account settings?

    No. The agent only has permission to run predictions.

  • 04

    Does the AI save the images it processes?

    No. The agent only reads the image data to generate a prediction.

  • 05

    Can I get the bounding box coordinates?

    Yes. The prediction tool returns the spatial data for detected objects.

  • More questions about Clarifai? See everything the Clarifai Connector can do

Connect Clarifai to Claude, Cursor, ChatGPT & more