> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mage.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Build and manage Mage data pipelines directly from your AI coding assistant.

**Mage Agent** connects your local development environment to your Mage Pro cluster through the [Model Context Protocol (MCP)](https://modelcontextprotocol.io). Your AI assistant can inspect, create, modify, and run Mage pipelines using natural language — and a built-in sync engine keeps local files in step with the cluster.

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/agent/quickstart">
    Get connected in under five minutes.
  </Card>

  <Card title="IDE Integration" icon="code" href="/agent/ide-integration">
    Connect Cursor, Claude Code, or Codex CLI.
  </Card>

  <Card title="Sync" icon="arrows-rotate" href="/agent/sync">
    Mirror your Mage project to local files.
  </Card>

  <Card title="MCP Tools" icon="wrench" href="/agent/mcp-tools">
    Full reference for every tool your AI can call.
  </Card>
</CardGroup>

## How it works

```
Your AI assistant  ──MCP──►  mage-agent  ──REST──►  Mage Pro cluster
      │                           │
      │                     sync engine
      └── local files ◄──────────┘
```

1. **MCP server** — mage-agent runs as a background stdio process. Your IDE or AI agent sends tool calls over JSON-RPC; mage-agent translates them into Mage Pro REST API requests.
2. **Skill resources** — the server exposes Markdown guidance files that give your AI assistant context on how to interact with pipelines, blocks, runs, and sync operations.
3. **Sync engine** — a separate CLI layer lets you pull the full remote project tree to a local directory, push local edits back, and resolve conflicts file by file.

## Supported clients

<CardGroup cols={3}>
  <Card title="Cursor" icon="arrow-pointer" />

  <Card title="Claude Code" icon="terminal" />

  <Card title="OpenAI Codex CLI" icon="terminal" />
</CardGroup>

Any MCP-compatible client that can launch a subprocess will work with the same configuration.