# Nyckel ML MCP > Nyckel ML connects your AI agent to advanced machine learning tools for automated data classification and semantic search. You can test custom models, classify text or images instantly, and find similar samples using natural language—all without writing a single line of integration code. It lets you manage the entire lifecycle of your ML assets right from your chat client. ## Overview - **Category:** developer-tools - **Price:** Free - **Tags:** machine-learning, classification, semantic-search, automated-labeling, predictive-modeling, data-tagging ## Description This MCP gives your AI agent access to professional machine learning capabilities. Instead of building complex APIs or running batch jobs, you simply ask your agent to classify data or search a gallery. Need to know if an uploaded image is a product or just clutter? You prompt the system, and it runs the appropriate ML function, giving you instant predictions along with confidence scores. It's useful for everything from content moderation to e-commerce research. If your team needs to build custom data workflows, this connection makes it possible. Your agent can list existing functions or look at training samples to check accuracy before making a prediction. When you connect this MCP via Vinkius, you get access to all these features through one conversational point. You’re doing deep ML work, but the interaction feels like just asking a smart teammate for an opinion. ## Tools ### annotate_ml_sample Assigns a specific label to an existing data sample in your training set. ### create_ml_sample Adds a brand new piece of raw data to be used as a training sample for your models. ### delete_ml_function Permanently removes an existing machine learning function from your account. ### get_ml_function Retrieves specific configuration and metadata details for a single ML function by its ID. ### get_account_info Fetches general profile and workspace information about your connected Nyckel account. ### invoke_ml_function Runs a specific, trained ML function against new data to get an instant classification or prediction score. ### list_ml_functions Lists all the machine learning functions currently defined within your account. ### list_ml_labels Retrieves a comprehensive list of every available label and category used by your ML models. ### list_ml_samples Shows you an overview of all the current training samples stored in your account's database. ### semantic_search Searches through your data gallery to find other samples that are conceptually similar to a provided input. ## Prompt Examples **Prompt:** ``` Classify this text: 'The delivery was very late and the food was cold' using function ID 'func_123'. ``` **Response:** ``` Invoking ML function func_123... The prediction for your text is 'Negative Sentiment' with a confidence score of 98.45%. Shall I check if there are similar historical samples in your database? ``` **Prompt:** ``` Search my product gallery for an image similar to 'https://example.com/shoe.jpg' using function 'func_search_99'. ``` **Response:** ``` Executing semantic search... I've found 3 semantically similar samples in your gallery. The top match is 'Running Shoe - Blue' (Confidence: 95.2%). Would you like the metadata for the matching samples? ``` **Prompt:** ``` List all the machine learning functions in my Nyckel account. ``` **Response:** ``` Retrieving ML functions... You have 4 active functions: 'Sentiment Classifier' (func_123), 'Product Search' (func_search_99), 'Logo Detector', and 'Spam Filter'. Which function would you like to inspect? ``` ## Capabilities ### Classify Content Send text or image URLs and receive instant predictions and confidence scores from your pre-trained machine learning functions. ### Perform Semantic Search Query existing search galleries to find samples that are conceptually similar, even if they don't contain the same keywords. ### Manage ML Functions List and retrieve detailed metadata for all machine learning functions available in your Nyckel account. ### Curate Training Data Upload new training samples, assign labels, or delete entire ML functions to refine model performance. ### Check Account Status Retrieve profile and workspace metadata for the authenticated Nyckel account. ## Use Cases ### Automating Content Screening A content moderator receives a flood of user messages and needs to classify sentiment and detect prohibited imagery. They simply tell their agent, 'Classify these 50 images using the Sentiment Classifier.' The agent executes invoke_ml_function for each image and returns a summary report with confidence scores. ### Debugging Model Performance A data scientist suspects one of their ML functions is biased. They use list_ml_samples to pull up the raw training data, then manually annotate_ml_sample on 20 records to check if human input aligns with the model's current labels. ### Finding Product Inspiration An e-commerce designer uploads a sketch of a new product and needs to see similar items sold previously. They prompt their agent, 'Find me products like this drawing,' triggering semantic_search against the entire product gallery. ### Checking Model Scope A developer joins a project mid-cycle and doesn't know what ML tools exist. They ask their agent to list all available functions using list_ml_functions, getting an instant overview of the entire system. ## Benefits - You get immediate, actionable predictions. Instead of waiting for a batch job or writing custom code to hit an endpoint, you simply ask your agent to run the classification via invoke_ml_function. - Your search is smarter. Using semantic_search means you don't have to guess keywords; your agent finds samples that are conceptually related to what you provide. - Data governance becomes easy. You can list all available labels using list_ml_labels, ensuring your classification process sticks to the defined schema every time. - You stay in control of your data pipeline. The MCP lets you monitor training progress by listing_ml_samples and manually assigning or updating tags with annotate_ml_sample. - Rapid prototyping is possible. AI developers can test multiple ML functions by list_ml_functions without ever leaving their chat environment. ## How It Works The bottom line is you talk to your AI agent like normal, and it handles all the complex data processing in the background. 1. Subscribe to this MCP and enter your unique Nyckel Client ID and Secret credentials. 2. Your AI client connects, establishing a secure link that grants access to all ML tools. 3. You issue a natural language command through your agent—for example, 'Classify this text' or 'Find samples similar to this image.'—and the MCP executes the required function. ## Frequently Asked Questions **How do I start classifying data using the Nyckel ML MCP?** You must first subscribe to this MCP and provide your client credentials. Once connected, simply ask your agent to classify content by referencing a specific function ID or label. **Can I find similar images without knowing keywords using Nyckel ML MCP?** Yes, use the semantic_search tool. This feature finds samples based on conceptual similarity rather than just matching text strings, making it perfect for product discovery or image recognition. **What is the difference between list_ml_labels and get_account_info?** list_ml_labels specifically gives you all available categories your models understand. get_account_info provides broader organizational metadata about your workspace and profile. **If I want to test a brand new ML model, what tool should I use in Nyckel ML MCP?** You should start by using list_ml_functions. This shows you all currently available models, helping you decide which function to invoke_ml_function for testing. **Does the Nyckel ML MCP help with data quality control?** Absolutely. You can monitor and improve your data by listing_ml_samples and using annotate_ml_sample to manually correct or add labels to existing training records.