amazon-mcp

MCP server for Amazon Selling Partner API (SP-API) and Advertising API — exposes orders, inventory, pricing, ads, reports, and related read paths to Claude, Cursor, and other MCP clients.

This connects your own Seller Central and Ads API data — an operations tool for sellers who already run a store, not a market or product-research tool. If you want category research, keyword mining, or competitor intel on products you do not yet sell, look elsewhere.

Audience: developers who already use MCP. You are expected to read tool docstrings, configure LWA credentials yourself, and run the server locally or on your own host.

30-second start

git clone https://github.com/coaxon/amazon-mcp.git
cd amazon-mcp
pip install -r requirements.txt
cp .env.example .env            # defaults: dry-run, no credentials
python -m amazon_mcp            # stdio → Claude Desktop / Cursor

Planned (not on PyPI yet): pip install amazon-mcp · uvx amazon-mcp

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "amazon-sp": {
      "command": "python3",
      "args": ["-m", "amazon_mcp"],
      "cwd": "/path/to/amazon-mcp",
      "env": {
        "AMAZON_MCP_DRY_RUN": "1",
        "AMAZON_MCP_DATA_DIR": "/path/to/amazon-mcp/data"
      }
    }
  }
}

HTTP transport (Cursor / remote client). Set AMAZON_MCP_API_KEY in server .env first.

AMAZON_MCP_TRANSPORT=streamable-http AMAZON_MCP_HOST=127.0.0.1 AMAZON_MCP_PORT=8780 python -m amazon_mcp

{
  "mcpServers": {
    "amazon-sp": {
      "url": "http://127.0.0.1:8780/mcp",
      "headers": {
        "Authorization": "Bearer ${AMAZON_MCP_API_KEY}"
      }
    }
  }
}

See claude_desktop_config.example.json.

Related MCP server: Amazon Seller MCP Server - DataDoe

Dry-run (no credentials)

Default AMAZON_MCP_DRY_RUN=1 serves bundled fixtures — no LWA app, no Seller Central auth.

cp .env.example .env
python -m amazon_mcp

In your MCP client, call:

amazon_health()
amazon_inventory(action="list_asins")
amazon_catalog(action="lookup", asin="B0POC00001")
amazon_orders(action="revenue_summary")
amazon_alerts(action="alert_config")

Fixture ASINs include B0POC00001, B0FIXTURE01, etc. Responses include "dry_run": true in metadata.

Live SP-API

Set in .env:

Step-by-step credential setup: docs/OPERATOR_QUICKSTART.md (§ SP-API 客户凭证申请指引).

Core tools (open source)

Domain tools follow amazon_<domain>(action="..."). Core includes:

Also in core: run_dag_plan / resume_dag_plan — three-phase SP-API executor (amazon_mcp/dag/); no Pro dependency.

Core vs Pro

Pro is a separate optional package (amazon-mcp-pro). Core detects it via importlib.util.find_spec("amazon_mcp_pro"). No license keys.

Without Pro, these return {"error": "pro_required", ...} (see Getting Pro for how to enable Pro today):

Pro adds: multi-tenant gateway, AlertEngine polling, Slack/Stripe integrations, scenario orchestration, cross-platform connectors. There is no self-serve install path yet — see Getting Pro below.

Getting Pro

Pro is not yet published as a standalone package. There is no PyPI release (public or private), so pip install amazon-mcp-pro will fail. Current options:

• Private deployment — we set up and host the full stack (multi-tenant gateway, AlertEngine, Slack/Stripe integrations, scenario orchestration) on your infrastructure or ours.

• Questions / business inquiries — open a GitHub Issue or email info@coaxon.me.

Environment reference

Full template: .env.example.

Deploy

Generic VPS install:

bash scripts/install.sh --install-dir /opt/amazon-mcp --systemd
bash scripts/verify_install.sh

Details: docs/DEPLOY_HANDBOOK.md, docs/RUNBOOK.md.

Development

AMAZON_MCP_DRY_RUN=1 python -m amazon_mcp
AMAZON_MCP_FORCE_CORE=1 python -m amazon_mcp   # simulate core-only in monorepo

Tests are being updated for the core/pro split; not required for trying dry-run locally.

License

MIT