# Weights & Biases MCP for AI Agents AI Agent Connect

> Weights & Biases MCP lets your AI agent navigate your experiment history, monitor live training runs, and manage model artifacts without you having to leave your chat window. It turns your machine learning research into a conversation, giving you instant access to metrics, configs, and data lineage.

## Overview
- **Category:** ai-frontier
- **Price:** Free
- **Endpoint:** https://edge.vinkius.com/vk_preview_I1Yu40vJsqaxsdrmMLW0Zf84kGzji4jYFvnrqvBH/ai-agent-connect
- **Tags:** machine-learning, experiment-tracking, mlops, model-versioning, data-artifacts, model-monitoring

## Description

Imagine you're deep in a training cycle and need to know if your latest model is actually converging or just overfitting. Instead of switching tabs, logging into a dashboard, and hunting through dozens of experiment folders, you just ask your agent. This Connector connects your Weights & Biases account directly to your workflow, making it possible to pull specific metrics, check hyperparameter configs, and see which artifacts are linked to which version of your data. It's about moving from manual dashboard navigation to just asking questions about your research. By using the Vinkius catalog to get this connected, you can treat your entire experiment history as a searchable database. You can see what's happening in your sweeps, find those specific model weights from last Tuesday, and get a summary of your project's progress in seconds. It removes the friction of wondering where you saw a specific result and replaces it with immediate access to your ML history. You don't have to hunt for information anymore; your agent brings the data to you as you work.

## Tools

### list_project_artifacts
Lists all datasets, models, and files stored within a specific project. Use this to audit your data lineage and file versions.

### list_wandb_projects
Shows all projects under your user account or team entity. This makes it easy to navigate your different research areas.

### list_project_runs
Lists every experiment run within a specific project. Use this to find the right run ID when you have dozens of experiments.

### list_project_sweeps
Shows all hyperparameter search sweeps currently happening or finished in a project. It lets you monitor optimization progress.

### get_run_details
Pulls the full summary, including metrics and config, for a specific experiment run. This helps you see exact results without opening a browser.

### list_project_reports
Displays all saved analysis reports and dashboards for a project. It helps you find collaborative documentation quickly.

## Prompt Examples

**Prompt:** 
```
Show me the latest runs in my 'llm-finetuning' project.
```

**Response:** 
```
I found the following recent runs in **llm-finetuning**:

| Run Name | Status | ID |
| :--- | :--- | :--- |
| `vibrant-sweep-1` | **Running** | `run_a1b2` |
| `crispy-forest-12` | **Finished** | `run_c3d4` |
| `solar-wind-15` | **Crashed** | `run_e5f6` |

Would you like to see the metrics for any of these?
```

**Prompt:** 
```
What were the hyperparameters for run 'opt-run-42'?
```

**Response:** 
```
Here are the configuration details for run **opt-run-42**:

*   **Learning Rate**: 0.001
*   **Batch Size**: 32
*   **Optimizer**: adam
*   **Epochs**: 50

It finished about 2 hours ago with a final accuracy of **0.942**.
```

**Prompt:** 
```
List the datasets in the 'vision-task' project.
```

**Response:** 
```
In the **vision-task** project, I found the following artifacts:

1.  **imagenet-subset** (Dataset, v3)
2.  **resnet50-weights** (Model, v5)
3.  **training-logs** (Artifact, v1)

Would you like to see the version history for any of these?
```

## Capabilities

### Pull specific run metrics
Get the latest accuracy, loss, and hyperparameter values for any specific experiment run.

### Browse project folders
See all projects and experiment folders in your account to find the right workspace quickly.

### Audit model artifacts
List datasets, weights, and other files to track your data lineage and dependencies.

### Track hyperparameter sweeps
Monitor the progress of automated optimization jobs to see how your search is performing.

### Access research reports
View saved analysis reports and collaborative dashboards to see team findings and documentation.

### Monitor live training
See the current status of active runs to catch crashes or performance issues early.

## Use Cases

### The 'Where was that?' scenario
A researcher needs to find the specific hyperparameters used in a run from three weeks ago. They ask the agent to list runs and then get the details for the one with the best accuracy.

### The 'Did it crash?' check
An engineer wants to know if the current training job is still running or if it hit an error. The agent checks the project runs and reports the status of the latest one.

### The 'Data Audit'
A team needs to verify which model weights are associated with the latest production dataset. The agent lists all artifacts to show the version history.

### The 'Sweep Summary'
A manager wants a quick overview of how the hyperparameter sweep is performing. The agent lists the sweeps and summarizes the progress.

## Benefits

- Stop digging through dashboards by using list_wandb_projects to jump straight to the right experiment folder in one step.
- Get instant answers on model performance by using get_run_details to pull accuracy and loss metrics without switching tabs.
- Keep your data lineage clean by using list_project_artifacts to see exactly which version of a dataset was used for a run.
- Monitor your hyperparameter tuning progress faster by using list_project_sweeps to see how your automated search is going.
- Save time on team updates by using list_project_reports to quickly surface collaborative documentation and saved analyses.

## How It Works

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

1. Subscribe to the Weights & Biases MCP on Vinkius.
2. Enter your W&B API Key and optional Base URL in your AI client settings.
3. Ask your agent to list your recent runs or get the config for a specific run to start managing your experiments.

## Frequently Asked Questions

**Can I use the Weights & Biases MCP to see my training progress?**
Yes, you can monitor live runs in real-time. Your agent can check the status of active jobs and report back on whether they are running, finished, or have crashed.

**How do I access my team's projects with this Connector?**
Once connected, your agent can list all projects within your team entity. You can then ask it to find specific project folders or reports shared by your colleagues.

**Can this Connector help me find old model weights?**
Yes, you can use it to list all artifacts within a project. This allows you to quickly identify specific model versions and their associated data lineage.

**Does the Weights & Biases MCP support hyperparameter sweeps?**
Yes, it can list all hyperparameter search sweeps. This makes it easy to see how your automated optimization jobs are progressing without manually checking the web dashboard.

**How do I get my W&B data into my AI client?**
You can connect the Weights & Biases MCP through the Vinkius catalog. Once linked, your AI client can access your experiment data using your W&B API key.

**Can I check the latest metrics for a specific ML run?**
Yes. Using the `get_run_details` tool, your AI agent can pull the latest logged metrics (like accuracy or loss) and hyperparameters for any specific run ID within your projects.

**Is it possible to list versioned datasets and models?**
Absolutely. The `list_project_artifacts` tool allows you to see all artifacts, including datasets and models, helping you track data lineage and versioning directly through conversation.

**Can I monitor hyperparameter search sweeps via chat?**
Yes. Use the `list_project_sweeps` tool to monitor automated optimization tasks. Your agent will return a list of sweeps in the project so you can track progress without leaving your workspace.