# DoorDash MCP Server

An MCP (Model Context Protocol) server that lets AI agents search restaurants, browse menus, compare prices, and manage your DoorDash cart — all without opening a browser or wasting tokens on HTML parsing.

## How It Works

Instead of using a browser-based AI agent (which eats context and tokens parsing HTML), this MCP server runs a lightweight background browser that handles authentication and API calls. Your AI agent gets clean JSON — no HTML, no screenshots, no wasted tokens.

Under the hood, it uses DoorDash's internal GraphQL API (reverse-engineered from web traffic) via a Playwright browser instance that maintains your session.

## Features

| Tool | Description |
|------|-------------|
| `login_check` | Check if your DoorDash session is active |
| `search_restaurants` | Search restaurants and food by keyword |
| `get_store_menu` | Get full menu with prices, deals, and badges |
| `add_to_cart` | Add items to your cart |
| `remove_from_cart` | Remove items from cart |
| `list_carts` | View all active carts |
| `order_history` | Get recent order history |

Deals and promotions (Buy 1 Get 1 Free, DashPass offers, etc.) are surfaced in both search results and menu items.

## Setup

### 1. Install

```bash
git clone <this-repo>
cd doordash-mcp
npm install
npx playwright install chromium
```

### 2. Configure Email

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

Edit `.env` and set your DoorDash account email:

```
DOORDASH_EMAIL=your-email@example.com
```

### 3. Login (one-time)

```bash
node login.js
```

This opens a browser, sends an OTP to your email/phone, and saves the session. You only need to do this once (or when your session expires).

**On headless Linux:** The script auto-starts a virtual display via Xvfb. Make sure `xvfb` is installed (`sudo apt install xvfb`).

### 4. Add to Claude Code

Add to `~/.claude/settings.json`:

```json
{
  "mcpServers": {
    "doordash": {
      "command": "node",
      "args": ["/absolute/path/to/doordash-mcp/mcp-server.js"]
    }
  }
}
```

Restart Claude Code to pick up the new server.

## Usage

Once configured, just talk to your AI agent naturally:

- *"Search for biryani near me"*
- *"Show me the Pizza Hut menu"*
- *"What's the cheapest dosa at Thanjai Restaurant?"*
- *"Add 2 Masala Dosas to my cart"*
- *"What did I order last time?"*
- *"Find me a burger place with deals"*

## How the Spy Tool Works

Want to discover new endpoints or debug? The `spy.js` script opens a browser and logs all DoorDash API traffic as you browse:

```bash
node spy.js
```

Browse DoorDash normally — search, view menus, add to cart. All API calls get logged to `logs/api-calls.jsonl`. Close the browser when done.

## Architecture

```
AI Agent  ──(MCP stdio)──>  mcp-server.js  ──(GraphQL via Playwright)──>  DoorDash API
                                 │
                            browser-data/    (persistent session cookies)
```

- **No headless HTTP**: Cloudflare blocks plain HTTP requests. The server uses a real browser (positioned off-screen on macOS, or via Xvfb on Linux)
- **Persistent session**: Login once, the browser profile in `browser-data/` keeps your cookies alive
- **Minimal tokens**: AI agents get structured JSON, never HTML

## Platform Notes

| Platform | How it runs |
|----------|------------|
| macOS | Browser window positioned off-screen (-32000, -32000) |
| Linux with display | Same as macOS |
| Linux headless (SSH) | Auto-starts Xvfb virtual display |

## Disclaimer

This project reverse-engineers DoorDash's internal web APIs for personal use. It is not affiliated with, endorsed by, or connected to DoorDash in any way. Use at your own risk — endpoints may change without notice, and automated access may violate DoorDash's Terms of Service.

## License

MIT