# H2O.ai MCP > H2O.ai controls your entire machine learning lifecycle directly from your AI agent. Use this MCP to audit model performance, track training jobs, and check the health of your cloud cluster without manually opening dashboards or running complex commands. ## Overview - **Category:** ai-frontier - **Price:** Free - **Tags:** machine-learning, model-lifecycle, data-frames, predictive-modeling, cluster-monitoring ## Description This connector lets you manage everything happening inside your H2O.ai instance using natural conversation. Instead of logging into a dashboard, you simply ask your agent for status updates. You can review existing machine learning models by listing them and verifying their performance metrics. Need to check the underlying data? Ask the MCP to list available structured datasets or retrieve specific columns from a frame. It even monitors long-running tasks; just query the jobs list to see if training is on track. Plus, you never have to worry about hardware limits again, because you can always ping root endpoints using `cloud_status` to verify memory utilization and overall cluster health. Connecting this MCP via Vinkius means your agent has instant access to all these deep ML operations, letting you orchestrate complex data science workflows in plain English. ## Tools ### list_frames Lists all structured datasets (frames) currently loaded into the H2O cluster. ### get_frame Retrieves specific column data from a named dataset frame. ### list_models Lists all machine learning models that have been saved and tracked in the cloud instance. ### get_model Retrieves detailed information, including performance metrics, for a specific named model. ### list_jobs Lists recent and ongoing tasks or training jobs running on the cluster. ### cloud_status Gathers hardware health metrics, including memory usage and operational status of the entire cloud cluster. ## Prompt Examples **Prompt:** ``` List all machine learning models in my H2O cluster ``` **Response:** ``` Retrieving models... I found 5 models including 'GBM_model_1', 'DeepLearning_v2', and 'AutoML_leader'. Would you like the detailed performance metrics for the AutoML leader model? ``` **Prompt:** ``` What is the current status of the H2O cloud cluster? ``` **Response:** ``` Checking cloud status... Your H2O cluster is 'Healthy' with 4 active nodes. Memory usage is at 35% (14GB used out of 40GB total). All hardware architecture endpoints are operational. ``` **Prompt:** ``` Show me the last 3 training jobs ``` **Response:** ``` Retrieving jobs... I found 3 recent tasks: 1. GBM Training (Completed, 10m ago). 2. XGBoost Grid Search (Running, 45% complete). 3. Data Parsing (Completed, 1h ago). I can provide more details for any of these jobs. ``` ## Capabilities ### Audit Model Inventory List, check details for, and verify the performance metrics of every machine learning model saved in your H2O cluster. ### Track Data Sources View structured datasets loaded into the cluster or retrieve specific dimensional data columns from a frame. ### Monitor Training Jobs Check the status and progress of queued or running model training jobs over time. ### Assess Cluster Health Get real-time diagnostics on the physical hardware, including memory usage and operational status of the cloud cluster. ## Use Cases ### Verifying Pre-Deployment Data Readiness A data scientist needs to ensure a new model can use the correct data fields. They ask their agent to `list_frames` to confirm the dataset exists, then use `get_frame` on that frame name to validate the precise column names before running training. ### Debugging Failed Live Models A product team notices a model's performance dip. They ask their agent to run `get_model` for that specific asset, which immediately returns detailed metrics and configuration blocks needed to diagnose the failure point. ### Checking Resource Limits Mid-Run An ML engineer is running a large training job. Before committing resources, they ask the agent to `cloud_status` to check current memory utilization and confirm there's enough overhead for the next task. ### Auditing Historical Runs A developer needs to know which models were trained last week. They use the MCP to `list_models`, filtering by date, then ask the agent to `list_jobs` to see the execution history for those specific model names. ## Benefits - You stop clicking through multiple dashboards to check status. With `cloud_status`, you get a single, immediate report on hardware health and memory usage for your entire H2O instance. - Model auditing becomes instant. Instead of manually scrolling version logs, use `list_models` or `get_model` to pull performance metrics and verify which models are deployed. - Data preparation is faster. Use `list_frames` to see what data is available in the cluster, then use `get_frame` if you need specific dimensional columns for a test run. - Tracking pipelines is straightforward. The `list_jobs` tool gives you a quick overview of all running training tasks and how far along they are. - The process moves from manual effort to conversation. By connecting this MCP via Vinkius, your agent handles the complex API calls behind the scenes so you just talk to it. ## How It Works The bottom line is that you get a single conversational entry point to manage complex data science operations. 1. Subscribe to this MCP and provide your H2O.ai Base URL. 2. Connect your agent (Claude, Cursor, etc.) using the provided credentials. 3. Start asking natural language questions like 'Show me all running jobs' or 'What is the memory usage?' ## Frequently Asked Questions **How do I check memory usage with H2O.ai MCP?** You run `cloud_status`. This tool gives you real-time diagnostics on hardware health, including how much memory is used and the total capacity of the cluster. **Can I list all my machine learning models using H2O.ai MCP?** Yes, use `list_models`. It pulls a comprehensive inventory of every model saved in your cloud instance so you know exactly what assets exist. **What is the difference between list_jobs and get_model?** `list_jobs` shows the history and current status of running tasks (like training runs). `get_model` provides the detailed metrics and configuration for a specific, already finished model asset. **How do I validate data columns using H2O.ai MCP?** First, use `list_frames` to see the available datasets. Then, specify which dataset you want and ask the agent to run `get_frame` to pull out specific column details. **Does H2O.ai MCP help with data schemas?** Absolutely. You can use `list_frames` and then `get_frame` to confirm the exact dimensional mapping and structure of your loaded datasets, ensuring schema integrity. **Can my agent list all data frames currently loaded in my H2O cluster?** Yes. Use the 'list_frames' tool. The agent retrieves the list of structured datasets securely loaded into memory, including their IDs and basic metadata, allowing you to browse available data flawlessly. **How do I check the progress of a model training job via chat?** Use the 'list_jobs' tool. Your agent will query the timeline nodes tracking all long-running tasks on the cluster, providing you with the current execution status and progress percentages synchronously. **Can I see the internal architecture and metrics of a model through the agent?** Absolutely. Use the 'get_model' tool with the specific model ID. The agent will fetch the detailed configuration blocks, exposing hyperparameters and performance metrics natively within your chat context.