amazon-in-mcp

A stdio MCP server for Amazon.in that drives a single, persistent, logged-in Playwright/Chromium session. It lets an MCP client (Claude Code, Claude Desktop, etc.) search products, read your orders, and analyze your order trends.

⚠️ Personal-use tool, unofficial. Not affiliated with, endorsed by, or associated with Amazon. It automates your own logged-in Amazon account via a browser — there is no official consumer API for order history, so it reads the same pages you see. Heavy automation can trip CAPTCHAs and brushes against Amazon's Terms of Service. Use it privately and responsibly; you are responsible for your own account and how you use it.

Tools

Related MCP server: Amazon MCP Server

Install as a Claude Desktop extension (MCPB) — easiest

A prebuilt MCPB bundle (.mcpb) is attached to each GitHub Release. This is the one-click path for Claude Desktop — no cloning or building.

1. Download amazon-in-marketplace-mcp.mcpb from the latest release.

2. Install it: open Claude Desktop → Settings → Extensions → drag the .mcpb file in (or double-click the file). Review the tools it exposes and click Install.

3. Configure (optional) in the extension's settings pane:

   • Run headless — on by default (no window). Turn off if order pages start showing CAPTCHAs.

   • Idle auto-close (ms) — how long the browser lingers before closing itself (default 45000; 0 = never).

4. Sign in — no terminal needed. Just ask Claude to run the login tool. A Chrome window opens at the Amazon.in sign-in page and stays open; sign in by hand (including OTP/CAPTCHA). When you see "Hello, <name>", ask Claude to run finish_login — it confirms, closes the window, and goes back to headless. Your session is saved to ~/.amazon-in-mcp/profile and reused on every later call.

5. Chromium requirement. The bundle ships the server + Node deps but not the Chromium binary (Playwright keeps that in a shared cache). If the login tool errors that a browser is missing, run once in a terminal: npx playwright install chromium

To build the .mcpb yourself: npm run build && npx @anthropic-ai/mcpb pack .

Install from Git

Clone the repo and build from source (there is no npm-registry package — install straight from Git):

# 1. Clone
git clone git@github.com:kausikram/amazon-in-marketplace-mcp.git
cd amazon-in-marketplace-mcp

# 2. Install deps (postinstall also downloads the Chromium browser binary)
npm install

# 3. Build the TypeScript
npm run build

# 4. Build only — no CLI login step needed. Sign in later from your MCP client
#    by asking Claude to run the `login` tool (opens a window), then
#    `finish_login`. (A CLI fallback still exists: `npm run login`.)

You can also install a specific commit or branch by cloning that ref, e.g. git clone -b main git@github.com:kausikram/amazon-in-marketplace-mcp.git. After npm run build, the runnable entry point is dist/index.js — that's the path you register with your MCP client (see Register below).

Requires Node 18+ and enough disk for the Chromium download (~150 MB).

Setup

cd amazon-in-mcp
npm install            # also runs `playwright install chromium`
npm run build

Signing in: you don't need the command line. Once the server is registered with your MCP client, ask Claude to run the login tool — a Chrome window opens at the sign-in page; sign in by hand, then run finish_login. The session is stored in a persistent profile at ~/.amazon-in-mcp/profile (override with AMAZON_MCP_PROFILE_DIR) and reused, so you sign in just once.

CLI fallback: npm run login does the same thing from a terminal if you prefer.

Environment variables

Register with Claude Code

Add to your .mcp.json (or claude mcp add):

{
  "mcpServers": {
    "amazon-in": {
      "command": "node",
      "args": ["/Users/kausikram/MCP_Servers/amazon-in-mcp/dist/index.js"],
      "env": {
        "AMAZON_MCP_HEADLESS": "0"
      }
    }
  }
}

Then, in a chat: "search amazon for a 100W usb-c charger", "list my orders from the last 3 months", "analyze my 2024 amazon spending."

How it works / where to fix things

• src/browser.ts — the single persistent Chromium context (profile reuse, stealth touches).

• src/auth.ts — sign-in / CAPTCHA detection; every order tool guards on it.

• src/parse.ts — all fragile DOM selectors live here. When Amazon changes markup, fix this one file.

• src/tools/* — the five tools.

• src/index.ts — MCP wiring over stdio.

Known limitations

• Order-history has no API; scraping breaks when Amazon changes HTML — update parse.ts.

• CAPTCHA/OTP require a manual npm run login again.

• Trend analysis derives from the order-list view (dates, totals, item titles), not deep per-item category data. Extend get_order + trends.ts for category-level analysis.

License

Released into the public domain under The Unlicense — the most lenient option available. Do anything you like with it: copy, modify, sell, redistribute, no attribution required, no warranty. See LICENSE.