# Comet ML MCP for AI Agents AI Agent Connect

> Comet ML lets you manage your machine learning lifecycle through your AI agent. Track model metrics, audit project workspaces, and inspect run parameters without switching tabs. It connects your experimental data directly to your chat interface for faster debugging and analysis.

## Overview
- **Category:** ship-it
- **Price:** Free
- **Endpoint:** https://edge.vinkius.com/vk_preview_C2PxqYqwffQi3V41WE8jlPynFSP048sA8SVR1PFT/ai-agent-connect
- **Tags:** mlops, experiment-tracking, model-evaluation, llm-monitoring, model-lifecycle, data-science

## Description

Imagine you're in the middle of a long training run and need to know if the loss curve is behaving or if your learning rate is actually hitting the target. Usually, this means jumping out of your editor, opening a browser, and digging through a dashboard to find the right data. This Connector changes that by letting you ask your AI agent for the status of any specific experiment using natural language. You can ask it to find a project from last week, pull the latest accuracy numbers, or check the exact hyperparameters you used for a specific model version. It brings your entire machine learning history into a single conversation. Whether you're trying to compare two different training loops or just need to verify that a production run finished successfully, you can do it by just asking. It's a way to keep your research moving forward without the friction of manual navigation or context switching. By adding this to your setup via the Vinkius catalog, you turn your AI into a specialized assistant that knows your data as well as you do. Instead of hunting through folders and tabs, you get direct answers about your model performance and configuration. This means you can spend less time auditing and more time actually refining your models.

## Tools

### list_workspaces
See all available routing spaces in your Comet ML account. This helps you find the right organizational area for your data.

### list_projects
Find specific projects within your designated Comet ML workspaces. Use this to see what research is currently active.

### list_experiments
Get a list of all logged experiments for a specific project. This is useful for seeing every training run you've conducted.

### get_experiment
Fetch the specific details and logs for a unique experiment ID. Use it to see the full history of a single run.

### get_experiment_metrics
Pull the latest numeric metrics for a specific experiment run. This gives you the most recent accuracy and loss data.

### get_experiment_params
View the hyperparameters and configuration details of an experiment. Use this to verify your learning rates and model settings.

## Prompt Examples

**Prompt:** 
```
Which experiment had the highest accuracy in the 'Vision-Task' project?
```

**Response:** 
```
I've analyzed the experiments in the **Vision-Task** project. Here are the top performers:

| Experiment ID | Accuracy | Loss | Status |
| :--- | :--- | :--- | :--- |
| `exp_v2_final` | **0.945** | 0.121 | Completed |
| `exp_v2_beta` | 0.921 | 0.145 | Completed |
| `exp_v1_test` | 0.889 | 0.180 | Completed |

The `exp_v2_final` run is currently your best performer.
```

**Prompt:** 
```
Show me the learning rate and batch size for experiment exp_998.
```

**Response:** 
```
Here are the hyperparameters for experiment **exp_998**:

*   **Learning Rate:** 0.001
*   **Batch Size:** 32
*   **Optimizer:** adam
*   **Model Architecture:** resnet50

Would you like me to check the metrics for this specific run as well?
```

**Prompt:** 
```
Are there any active runs in the 'Production-Alpha' workspace?
```

**Response:** 
```
Checking the **Production-Alpha** workspace now...

Yes, there is one active run:
*   **Run ID:** `prod_run_772`
*   **Status:** In Progress
*   **Current Epoch:** 45
*   **Current Loss:** 0.12

The run is still active and performance appears to be trending upwards.
```

## Capabilities

### Audit live training metrics
Check the real-time performance of your active training runs.

### Inspect experiment hyperparameters
See the exact settings like learning rate or batch size for any run.

### Navigate project workspaces
Move between different organizational areas and research projects.

### Retrieve specific run logs
Get the full history and metadata for a single experiment.

### Filter experiments by tags
Find specific runs based on custom labels or metadata.

## Use Cases

### Comparing model accuracies
A researcher wants to know which of 50 runs had the best accuracy. They ask the agent to list all experiments and sort them by the best metric.

### Debugging a failed run
An engineer needs to see the parameters of a failed run. They ask the agent to get the params for the last failed experiment ID.

### Onboarding a new team member
A team lead wants to show a new dev where the 'NLP-v2' project lives. They ask the agent to list workspaces and projects.

### Production monitoring
An MLOps lead wants to check if the nightly training finished. They ask the agent to check the status of the latest run.

## Benefits

- Stop switching tabs by using `list_projects` to find research directly in your chat interface.
- Get instant feedback on training progress by pulling data with `get_experiment_metrics` to see current loss and accuracy.
- Verify your model's setup quickly using `get_experiment_params` to check learning rates and optimizer settings.
- Audit your entire history with `list_experiments` to see every run you've ever logged for a specific project.
- Navigate complex organizational structures using `list_workspaces` to find team data across different environments.

## How It Works

The bottom line is you get a conversational interface for your entire machine learning experiment history.

1. Subscribe to the Comet ML MCP on Vinkius.
2. Add your Comet ML API Key from your account settings.
3. Ask your AI agent to find projects or pull metrics.

## Frequently Asked Questions

**Can the Comet ML MCP help me see my training metrics?**
Yes, it lets your AI agent pull real-time accuracy, loss, and other numeric data from your Comet runs directly into your chat.

**How do I find a specific project using Comet ML?**
You can just ask your agent to list your projects or search for a specific one by name, and it will find the correct project ID for you.

**Can I use this to check my hyperparameters?**
Absolutely. You can ask the agent to retrieve the exact configuration for any specific experiment, including learning rates and batch sizes.

**Does the Comet ML MCP work with my current AI client?**
It works with any MCP-compatible client, including Claude, Cursor, and Windsurf.

**Can I see all my workspaces at once?**
Yes, the agent can list all your Comet workspaces to help you navigate different teams or projects quickly.

**Is this for running my models?**
No, this is for auditing and inspecting experiments you've already logged in Comet ML. It doesn't execute the training code itself.

**Can my agent retrieve real-time metrics from an active ML run?**
Yes. Use the 'get_experiment_metrics' tool with the experiment key. The agent will pull the latest numeric logged endpoints, allowing you to monitor loss, accuracy, and other custom metrics as they are generated.

**How do I audit the parameters used in a specific experiment?**
Provide the experiment key to your agent. The 'get_experiment_params' tool extracts all logged ML properties, helping you verify hyperparameters like learning rates, batch sizes, and model architectures.

**Can I see a list of all experiments within a specific project?**
Absolutely. Use the 'list_experiments' tool with the project ID. Your agent will surface all ML runs within that project, including their status and metadata, so you can quickly identify the results you need.