Nyckel ML MCP. Classify Data and Manage Models via Agent Chat

You're connecting your AI agent directly to a Nyckel ML MCP Server. This gives your client hands-on access to machine learning workflows—it handles automated classification and semantic searching right from natural conversation. You don't write custom API calls or build complex integrations; you just tell the agent what job needs doing.

Classifying Data with Trained Models

You can pass data—whether it’s text or image URLs—to a specific, trained function using invoke_ml_function. The system runs the model and immediately returns an instant prediction along with a confidence score. This is how you classify new inputs against existing ML models.

Finding Similar Data Through Search

Don't rely on keyword matching. You run semantic_search to query your sample gallery based on conceptual meaning. This function finds samples that mean the same thing as your input data, even if they use totally different words or phrasing.

Managing ML Functions and Accounts

You need an overview of what's running? You can list every active machine learning function in list_ml_functions within your account namespace. If you want deep details on one specific model, use get_ml_function to fetch its exact configuration by ID. For general setup checks, get_account_info retrieves basic metadata about your entire Nyckel ML account.

Building and Controlling Training Samples

The server gives you full control over the data used for training. You can start by listing existing samples using list_ml_samples to check all current records and associated metadata. Need more training material? Use create_ml_sample to add a brand-new data point into your gallery. Once the sample exists, you'll use annotate_ml_sample to assign it a specific classification label, which directly improves your model accuracy.

You also have the ability to clean up by removing old models using delete_ml_function.

Defining and Discovering Labels

The agent handles all the necessary data organization for you. To see what categories are available across your entire setup, call list_ml_labels. This function provides a complete list of label definitions (the categories) used by your models. You'll never have to guess what labels exist.

This tool set lets you build sophisticated ML pipelines without writing boilerplate code. It keeps everything—from initial data entry to final prediction scoring—within the agent's command structure.