---
title: Auto Deployer
emoji: 🚈
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
license: apache-2.0
short_description: From CSV to Deployed ML API in 30 seconds.
tags:
  - building-mcp-track-enterprise
---


# Auto-Deployer MCP Server

**Auto-Deployer** is a Model Context Protocol (MCP) server that gives AI agents (like Claude, Cursor, or Windsurf) the ability to **train and deploy machine learning models** directly from a conversation.

It bridges the gap between an AI agent's reasoning capabilities and heavy-duty ML infrastructure. By connecting this server, your agent gains tools to:
1.  **👀 Analyze** CSV datasets with comprehensive insights.
2.  **🚈 Train** models (Classification, Regression, Time Series) on serverless infrastructure.
3.  **🚀 Deploy** those models as production-ready APIs instantly.

> Check out the social media post with the video demo: https://www.linkedin.com/feed/update/urn:li:ugcPost:7399114210342903809/

> Here is the youtube video: https://www.youtube.com/watch?v=F1Xxvz95N1E

---

## 🛠️ How It Works

This project uses a streamlined architecture for maximum efficiency:

1.  **The Interface (Gradio MCP)**:
    *   The `app.py` script runs a Gradio application that serves as the **MCP Server**.
    *   It exposes **4 core tools** as MCP functions that AI agents can discover and invoke.
    *   Features **smart file handling** that seamlessly processes uploads, URLs, and local paths.
    *   Provides both **web interface** for manual use and **MCP interface** for agent integration.

2.  **The Engine (Modal)**:
    *   The `modal_backend.py` runs on **Modal's serverless cloud platform**.
    *   When agents request model operations, the Gradio server triggers Modal functions.
    *   Modal automatically scales compute resources, processes data, trains models, and serves them as production APIs.
    *   Models are stored persistently and available via high-performance endpoints.

### 🏗️ Simplified Architecture
- **4 Core Tools Only**: Eliminates complexity with focused functionality
- **Unified Interface**: Same tools work for web users and AI agents
- **Smart Input Detection**: Handles files, URLs, and uploads automatically
- **Serverless Scaling**: Pay only for compute when you use it
- **Production Ready**: Instant API deployment with comprehensive monitoring

---

## 🚀 Usage Guide

This MCP server is hosted on Hugging Face Spaces. You do not need to run any code locally to use it, other than configuring your AI agent client.

### Prerequisites
*   **Claude Code** (or another MCP-compliant client).
*   **UV**: Required to run the MCP tools (`pip install uv`).

---

## 🔌 Connecting to Claude Code

To connect your agent to the live server, add the following configuration to your `.claude.json`:

```json
{
  "mcpServers": {
    "auto-deployer": {
      "type": "http",
      "url": "https://mcp-1st-birthday-auto-deployer.hf.space/gradio_api/mcp/"
    },
    "upload_files_to_gradio": {
      "command": "uvx",
      "args": [
        "--from",
        "gradio[mcp]",
        "gradio",
        "upload-mcp",
        "https://mcp-1st-birthday-auto-deployer.hf.space",
        "C:\\Path\\To\\Your\\Data"
      ]
    }
  }
}
```

**Important Notes:**
*   **`auto-deployer`**: Connects to the live Auto-Deployer MCP server.
*   **`upload_files_to_gradio`**: This helper tool allows the agent to "upload" local files from your computer to the remote server for processing.
*   **Action Required**: Replace `C:\\Path\\To\\Your\\Data` with the absolute path to the folder containing your CSV files.

---

## 💬 Example Prompts

Once connected, you can talk to your agent naturally. Here are some workflows you can try:

### 👣 The Step-by-Step Approach
*Best for understanding the data first.*

**Step 1: Upload and Analyze**
> "Upload my local data folder using the `upload_files_to_gradio` tool. Then, analyze `housing_prices.csv` and tell me about the dataset shape, column types, and any missing values."

**Step 2: Training**
> "Train a regression model to predict 'price' using the housing data. Focus on features like 'sqft_living', 'bedrooms', and 'location'."

**Step 3: Deployment**
> "Deploy the trained model and provide me with the API endpoint and Python code examples for making predictions."

### ⚡ The One-Click Approach
*Best for rapid deployment.*

**Complete Pipeline**
> "Upload `customer_churn.csv` and auto-deploy a classification model to predict 'Churn' status. Give me the full report with performance metrics and API access."

### 🔧 Advanced Workflows

**Custom Model Training**
> "Analyze this medical dataset, then train a classification model for 'HeartDisease' with 80/20 train-test split and provide detailed performance metrics."

**Batch Processing**
> "I have multiple datasets in my folder. Upload all CSV files and analyze each one, then train models for the datasets that have good data quality."

---

## 📦 Available Tools

The server provides **4 core tools** that work seamlessly with both file uploads and URLs:

### 🔧 Core Tools

| Tool Name | Description | Input | Output |
| :--- | :--- | :--- | :--- |
| `analyze_data_tool` | 📊 Analyze CSV datasets and provide comprehensive statistical metadata | CSV file or URL | JSON analysis with shape, columns, data types, and missing values |
| `train_model_tool` | 🚈 Train production-ready ML models on serverless infrastructure | CSV file/URL, target column, task type | JSON with model ID and performance metrics |
| `deploy_model_tool` | 🚀 Deploy trained models to live production API endpoints | Model ID | Markdown with API URL and usage examples |
| `auto_deploy_tool` | ⚡ Complete end-to-end pipeline (Analyze → Train → Deploy) in one click | CSV file/URL, target column, task type | Comprehensive deployment report with insights |

### 🎯 Task Types Supported
- **Classification**: Predict categorical outcomes (e.g., spam/not spam, disease/no disease)
- **Regression**: Predict continuous numerical values (e.g., house prices, temperature)
- **Time Series**: Forecast time-based data patterns and trends

### 📋 Input Flexibility
All tools accept multiple input formats:
- **Local CSV files**: Direct file uploads through the web interface
- **File URLs**: HTTP/HTTPS links to CSV files
- **File paths**: Local file system paths
- **Gradio file objects**: Seamless integration with web uploads

### 🚀 Key Features
- **Serverless Processing**: Runs on Modal's serverless infrastructure for scalability
- **Smart File Handling**: Automatic detection and processing of different input types
- **Comprehensive Analytics**: Detailed dataset insights and model performance reports
- **Production APIs**: Instant deployment with ready-to-use code examples
- **Multi-format Support**: Works with CSV files from various sources