# azure-adf-mcp

An [MCP (Model Context Protocol)](https://modelcontextprotocol.io) server that exposes **Azure Data Factory** operations as tools any LLM can call — trigger pipelines, monitor runs, inspect datasets, and get factory health summaries through natural language.

---

## Demo

> *"Trigger the CopyRawToStaging pipeline and tell me when it's done."*

Claude calls `trigger_pipeline`, gets back a run ID, polls `get_run_status`, and reports back — all without leaving the chat.
![Pipeline run 24hr view](images/trigger_pipeline.png)
![Pipeline run 24hr view](images/success.png)

> *"can you give me list of all pipline runs in last 24 hours."*
![Pipeline run 24hr view](images/pipeline_run_24_hr.png)

---

## Tools Exposed

### Pipelines

| Tool | Description |
|---|---|
| `list_pipelines` | All pipelines with activity breakdown |
| `get_pipeline_definition` | Full definition of a pipeline — activities, parameters, variables, and dependency chain |
| `create_pipeline` | Create a new empty pipeline (add activities in ADF Studio afterwards) |
| `trigger_pipeline` | Start a named pipeline run |
| `get_run_status` | Poll a run by ID |
| `list_recent_runs` | Recent runs with status summary (filterable by pipeline and time window) |
| `list_activity_runs` | Individual activity results within a run — pinpoint which activity failed and why |
| `cancel_run` | Cancel an in-progress run |

### Triggers / Schedules

| Tool | Description |
|---|---|
| `list_triggers` | All triggers with runtime state and linked pipelines |
| `get_trigger_status` | Current state and configuration of a specific trigger |
| `create_schedule_trigger` | Create a recurring schedule trigger (Minute / Hour / Day / Week / Month) |
| `start_trigger` | Activate a trigger so it starts firing on its schedule |
| `stop_trigger` | Deactivate a trigger |

### Datasets & Linked Services

| Tool | Description |
|---|---|
| `list_datasets` | All datasets and their linked services |
| `list_linked_services` | All storage/DB/API connections |

### Factory

| Tool | Description |
|---|---|
| `get_factory_summary` | High-level factory health overview |

---

## Requirements

- Python 3.9+
- An Azure subscription with a Data Factory instance
- [Azure CLI](https://learn.microsoft.com/en-us/cli/azure/install-azure-cli) installed and authenticated
- [Claude Desktop](https://claude.ai/download) (or any MCP-compatible client)

---

## Setup

### 1. Clone and install

```bash
git clone https://github.com/nachiketap11/azure-adf-mcp
cd azure-adf-mcp

python3 -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate

pip install -r requirements.txt
```

### 2. Configure environment

```bash
cp .env.example .env
```

Edit `.env` with your values:

```env
AZURE_SUBSCRIPTION_ID=your-subscription-id
AZURE_RESOURCE_GROUP=your-resource-group
AZURE_FACTORY_NAME=your-factory-name
```

### 3. Authenticate

```bash
az login
```

The server uses `DefaultAzureCredential` — Azure CLI login is all you need locally.

### 4. Test the server

```bash
python3 src/server.py
```

You should see: `Starting Azure ADF MCP Server...`

---

## Connect to Claude Desktop

Add to `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows):

```json
{
  "mcpServers": {
    "azure-adf": {
      "command": "/absolute/path/to/.venv/bin/python3",
      "args": ["/absolute/path/to/azure-adf-mcp/src/server.py"],
      "env": {
        "AZURE_SUBSCRIPTION_ID": "your-subscription-id",
        "AZURE_RESOURCE_GROUP": "your-resource-group",
        "AZURE_FACTORY_NAME": "your-factory-name"
      }
    }
  }
}
```

Restart Claude Desktop. You should see the 🔨 icon in the chat input bar with `azure-adf` tools listed.

---

## Project Structure

```
azure-adf-mcp/
├── src/
│   ├── server.py           # MCP server — tool definitions + routing
│   └── azure_client.py     # Azure SDK wrapper with clean JSON outputs
├── .env.example
├── requirements.txt
└── README.md
```

---

## Extending This Project

- Wrap with FastAPI to expose as an HTTP MCP server for remote deployment
- Add Pytest fixtures with mocked Azure SDK responses for CI
- Deploy to Azure Container Apps for persistent hosting

---

## Auth Reference

`DefaultAzureCredential` tries credentials in this order:

1. Environment variables (`AZURE_CLIENT_ID` / `AZURE_TENANT_ID` / `AZURE_CLIENT_SECRET`)
2. Workload identity (AKS)
3. Managed identity
4. Azure CLI (`az login`) ← recommended for local development
5. VS Code credentials

---

## License

MIT