# DVC MCP for AI Agents MCP

> DVC MCP connects your AI agent directly to your DVC Studio account for ML experiments. Stop clicking through dashboards and start asking natural language questions about model runs, project history, and data metrics. Audit projects, track views, and manage the entire lifecycle of your machine learning models via conversation.

## Overview
- **Category:** developer-tools
- **Price:** Free
- **Tags:** machine-learning, version-control, experiment-tracking, data-pipelines, model-management, git-workflow

## Description

Managing large-scale ML projects usually means jumping between a dozen different tabs: the dashboard, the Git repo, the metric logging service. It's slow, tedious, and prone to human error.

This MCP changes that. You connect your DVC Studio credentials once, and your AI client gets direct access to your entire data versioning workflow. Instead of manually navigating complex project structures or searching through log files for a specific accuracy score, you just ask your agent what you need.

You can tell it to list all active projects, check the history of model runs, or pull up structural details about dashboard views—all in plain English. It’s like having an expert ML Ops engineer sitting next to you, ready to answer any question about project data and versioning without ever leaving your chat window. This capability is available through Vinkius, making it easy to connect this core function into whatever AI client you already use.

## Tools

### list_views
Lists all defined dashboard views currently available in your DVC Studio account.

### get_view
Retrieves the detailed configuration and structural settings for a specific dashboard view.

### get_user
Returns basic profile information about the connected DVC Studio user account.

### list_projects
Retrieves a list of all registered ML projects (organizational workspaces) managed by your DVC Studio account.

### get_project
Fetches detailed metadata and status for one specific project identifier.

### list_experiments
Lists all recorded model experiments, showing key identifiers and run statuses within a given project.

## Prompt Examples

**Prompt:** 
```
What were the top 3 projects I ran last month?
```

**Response:** 
```
**Project Summary: Last 3 Months**

| Project Name | Best Metric (Accuracy) | Run ID |
| :--- | :--- | :--- |
| Credit-Scoring-Model | 0.942 | exp-a1b2 |
| Image-Classification-V2 | 0.875 | exp-c3d4 |
| NLP-Pipeline | 0.910 | exp-x9yz |

*Need to see the specific metrics array for any of these? Just ask.*
```

**Prompt:** 
```
Can you list all dashboard views and what they contain?
```

**Response:** 
```
You have 2 active dashboard layouts:

1. **Main-Metrics-Grid**: Displays primary KPIs (Accuracy, F1 Score) across a grid format.
2. **Validation-Plots**: Shows detailed plot structures for ROC curves and confusion matrices. 

You can get the full settings for either one if you want to modify it.
```

**Prompt:** 
```
What are my current user permissions in DVC?
```

**Response:** 
```
**DVC Studio User Profile**

*   **User:** Jane Doe (janedoe@corp.com)
*   **Role:** Data Scientist
*   **Scopes:** Read/Write access for all projects under the 'Finance' organization scope.
*   **Token Validity:** Active, expires 2026-12-31.
```

## Capabilities

### List all dashboard views
Retrieves a list of defined UI configuration layouts within your DVC Studio workspace.

### Retrieve specific view details
Fetches the structural settings and configurations for a single, chosen dashboard view.

### Get user profile information
Retrieves basic metadata about the authorized user account connected to DVC Studio.

### List all active projects
Provides a list of registered organization workspaces available within your DVC Studio environment.

### Get specific project details
Retrieves the full metadata and current status for an individual, specified ML project.

### List all model experiments
Generates a list of completed or running machine learning experiment runs tied to a specific project.

## Use Cases

### Finding the Best Model Run from Last Quarter
A data scientist needs to know which specific run achieved the highest accuracy for a fraud detection model. Instead of opening the dashboard and filtering by date, they ask their agent to list experiments, providing IDs and peak metrics immediately.

### Verifying Project Scope Before Development
An ML engineer starts a new task but needs confirmation that all required data sources are accounted for. They ask the agent to list projects across their organization to verify repository connections against internal team mappings, preventing build failures.

### Understanding Dashboard Configuration Changes
A team lead takes over a project and needs to know what metrics were tracked previously. Instead of clicking through the dashboard settings repeatedly, they ask their agent to list views and retrieve detailed configurations for all existing dashboards.

### Auditing Compliance and Access Rights
A DevOps engineer must confirm that only authorized personnel have access to sensitive model data. They use the agent to get user profile information and audit project metadata to verify current permissions against security guidelines.

## Benefits

- Instead of navigating through the DVC Studio UI to find model runs, you simply ask your agent to list experiments. This gives you instant access to run IDs, completion statuses, and performance summaries.
- You don't have to guess which dashboard layout is correct. Use the `list_views` tool to see all active views, then use `get_view` to pull up detailed structural settings for a specific one.
- Audit your entire ML portfolio easily. You can list projects and get full metadata on any workspace using `get_project`, helping you quickly verify if a dependency exists before starting work.
- Deep dive into performance metrics without manual logging searches. Your agent lets you request complex structural arrays defining exactly which metrics were captured during specific experiment epochs.
- Keep track of who's doing what. Get the current user profile using `get_user` to verify permissions and identify the authorized token holder when collaborating with a team member.
- When troubleshooting, quickly see all possible model experiments by calling `list_experiments`, giving you an immediate overview of the entire project history.

## How It Works

The bottom line is, you talk naturally about complex ML data versioning tasks, and this MCP handles the technical calls behind the scenes.

1. Subscribe to this MCP and provide your DVC Studio Client Access Token. This token grants the AI client permission to read your ML project data.
2. Tell your agent what you need, for example: 'Show me all projects I've set up.' The agent translates that request into a structured query for your DVC account.
3. The system executes the query and returns the specific, requested information—like a list of model runs or project metadata—directly back to your chat interface.

## Frequently Asked Questions

**How does DVC MCP help me track my model experiments?**
It lets you use natural language to audit your entire experiment history. Instead of clicking through dashboards, you can ask for specific metrics arrays or list all runs just by talking to your AI client.

**Can I find out what projects my team has set up?**
Yes. You simply ask the MCP to list all active projects. It gives you a clear overview of every organizational workspace, helping you manage dependencies and understand the scope of work.

**Is this better than just using the DVC Studio web interface?**
It's faster because it eliminates clicks. Instead of navigating multiple menus to find a specific project or view, your agent retrieves that data directly into the chat window in seconds.

**What kind of information can I get about dashboard views?**
You can list all available views and retrieve their structural settings. This is great for checking if a metric was tracked correctly or verifying which widgets are active on any given board.

**How do I verify my permissions using DVC MCP?**
If you need to check who has access or what scopes your token covers, you ask the agent for user profile information. This gives you a quick audit of authorized roles and tokens.