Yango Tech Retail MCP

MCP server for the Yango Tech Retail (grocery platform) B2B API: create and track orders, browse the product catalog, manage prices, discounts and stock levels across darkstores — from Claude, Cursor, Codex and other AI clients, in natural language.

The server wraps the retailer-facing API at api.retailtech.yango.com (the one the official yango-tech-grocery-client Python library speaks). The assistant assembles request bodies, follows pagination cursors, checks order states and receipts, and updates prices and stocks for you.

Quick start

1. Get a Bearer token from Yango Tech for your retailer account (via your Yango Tech integration manager — there is no self-service token portal).

2. Add the server — for example, in Claude Code (other clients):

   claude mcp add yango-retail \
     -e YANGO_RETAIL_TOKEN=your_token \
     -- npx -y mcp-yango-retail

3. Ask the assistant: "List our stores and show the current stock of product 4607034171438 in each."

Related MCP server: Amazon Seller MCP Server

What it can do

• Orders: create_order, cancel_order, get_order (details), get_orders_state (batch tracking), query_order_events (event feed: new orders, state changes, receipts), get_receipt (fiscal receipts).

• Catalog: get_stores (darkstores), query_products (cursor-based catalog feed), create_products (upsert, ≤100 per call).

• Pricing: query_price_lists, get_prices, set_prices (≤100 per call), create_discounts (≤100 per call).

• Stocks: query_stocks (feed across stores), update_stocks (modify or initialize, ≤1000 lines per call).

• Escape hatch: raw_request — any other /b2b/v1/* endpoint (orders/update, VAT, price-list links, 3PL delivery integration, …) with an SSRF guard on the path.

Resilience: per-request timeout (AbortController), automatic retries with exponential backoff (429 always; 5xx/network errors only for read calls — writes are never replayed), and error messages that carry the x-yatraceid/x-yarequestid headers Yango Tech support asks for.

Example prompts

• "Create order ORD-1042 for store LAVKA-3: 2 × product 123 at 150.00, cash on delivery."

• "What state are orders ORD-1040 through ORD-1042 in?"

• "Pull the full product catalog and list items missing a barcode."

• "Set the price of product 123 to 99.90 in price list default and check it back."

API access

• Backend: https://api.retailtech.yango.com, all endpoints under /b2b/v1/*, every call is a POST with a JSON body.

• Auth: Authorization: Bearer <token> on every request. Tokens are issued by Yango Tech per retailer account; treat them as secrets.

• There is no public first-party docs portal; the official Python client is the de-facto spec this server follows (see docs/TOOLS.md for known gaps, e.g. undocumented discount payload keys).

Installation

claude mcp add yango-retail \
  -e YANGO_RETAIL_TOKEN=your_token \
  -- npx -y mcp-yango-retail

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "yango-retail": {
      "command": "npx",
      "args": ["-y", "mcp-yango-retail"],
      "env": {
        "YANGO_RETAIL_TOKEN": "your_token"
      }
    }
  }
}

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in the project):

{
  "mcpServers": {
    "yango-retail": {
      "command": "npx",
      "args": ["-y", "mcp-yango-retail"],
      "env": {
        "YANGO_RETAIL_TOKEN": "your_token"
      }
    }
  }
}

Add to .vscode/mcp.json — note the key is servers, not mcpServers:

{
  "servers": {
    "yango-retail": {
      "command": "npx",
      "args": ["-y", "mcp-yango-retail"],
      "env": {
        "YANGO_RETAIL_TOKEN": "your_token"
      }
    }
  }
}

⚠️ The token is stored in plain text in the client's config file — restrict access to it.

Configuration

Requirements

• Node.js 20+.

• A Yango Tech Retail Bearer token.

Limitations

• This is a write API: create_order, cancel_order, create_products, set_prices, create_discounts and update_stocks change real data. Tools carry MCP annotations (read-only / write / destructive) so MCP clients can gate them.

• The official client self-throttles at 5 requests/second per (token, endpoint) — treat that as the safe budget; the server retries 429 with backoff but does not rate-limit proactively.

• Batch limits per request: 100 products / prices / discounts, 1000 stock lines.

• No delete endpoints exist for products, price lists, prices or discounts.

Documentation

• docs/TOOLS.md — every tool, response formats, known API gaps.

• docs/DEVELOPMENT.md — building, testing, telemetry.

• docs/PUBLISHING.md — release and registry checklist.

Telemetry

The server sends anonymous usage pings (event names, tool names, versions — never tokens, arguments or data). Opt out with ASKADS_TELEMETRY=0. Details in docs/DEVELOPMENT.md.

Support

Questions and issues: GitHub Issues or Telegram @gistrec.

License

MIT