Yango Tech Retail MCP (mcp-yango-retail) is an MCP server used to connect AI applications to retailer accounts in Yango Tech Retail. Through natural language, users can inspect stores, orders, products, prices, and inventory, and can also create orders and update account data when needed.
[Purpose] This server targets the retailer B2B API for grocery retail and dark stores, not the marketplace seller portal, taxi service, or Yango Delivery. It enables AI clients to operate retail data in a near-conversational way, and to clearly distinguish read-only, write, and potentially destructive operations before executing writes.
[Main capabilities] The server provides a total of 16 tools: 9 read-only tools, 5 write tools, and 2 potentially destructive tools, covering stores, catalog, pricing, inventory, orders, and receipts.
- Stores and product catalog: list stores and their id, status, location, address, and name (if available); browse products by cursor, view status, category, localized name, barcode, and custom attributes; create or update up to 100 products in a single request, with product records upserted rather than added as duplicates.
- Prices and discounts: list price lists and read product prices from one or more price lists; compare the price of the same product across different price lists; set up to 100 prices in a single request, using decimal strings such as "150.00"; after confirming the expected field structure, create up to 100 store-specific discounts.
- Inventory: read inventory for each store, including product, quantity, and shelf type; update up to 1,000 inventory rows for one store at a time; use initialize for the first inventory load and modify for daily updates.
- Orders and receipts: create orders after collecting store, products, quantities, prices, delivery details, and payment type; read order details and check the current status of multiple orders at the same time; follow the order event stream to get new orders, status changes, and issued receipts; read fiscal receipts by order id or receipt id; cancel orders after checking the current status and cancellation reason.
- Extended API: raw_request covers /b2b/v1/* methods for which no dedicated tool is provided, including order updates, VAT data, price list links, picking, logistics, and 3PL delivery operations. It can modify real account data and is intended for technical users who understand the upstream API.
[Usage] Requires Node.js 20+, a Yango Tech Retail account and a retailer Bearer token. The token must be obtained from the Yango Tech integration manager. After adding the server to an AI application, it is recommended to start with read-only requests, for example "List our stores and show the inventory of product [product ID] in each store". Configuration for different clients:
- Generic command-line launch: npx -y mcp-yango-retail@latest
- In a certain type of AI application: open Settings → MCP servers, select Add server, select STDIO, enter the launch command npx -y mcp-yango-retail@latest and the YANGO_RETAIL_TOKEN environment variable, save and restart.
- In Claude Desktop: the current official path is Settings → Extensions; for builds that still support local configuration, a JSON stdio configuration can be used as a fallback, saved on macOS to ~/Library/Application Support/Claude/claude_desktop_config.json and on Windows to %APPDATA%\Claude\claude_desktop_config.json.
- In Cursor: add on macOS/Linux to ~/.cursor/mcp.json and on Windows to %USERPROFILE%.cursor\mcp.json.
- In VS Code: run MCP: Open User Configuration and add the configuration, then check the server via MCP: List Servers. Connection check methods and the MCP documentation for Codex and Claude Code also provide reference paths.
[Data and Security] The server runs locally and sends retailer data directly to the configured Yango Tech Retail API host. The production API host is https://api.retailtech.yango.com。每个 API calls are all POST, with the JSON request body under /b2b/v1/*, including read operations. The Bearer token is only attached to requests to that host, and raw_request only accepts relative paths and rejects paths that resolve to other origins. By default, the server sends anonymous technical events to usage.gistrec.cloud, including server startup, the names of tools invoked, and a fixed reason code on startup failure; events include a random installation id, package version, AI client name and version, Node.js version, and operating system. The Bearer token, retailer data, tool parameters, and prompts are not read or sent. Telemetry has a two-second timeout and does not block tool calls. Telemetry can be disabled via configuration, with the implementation in src/telemetry.ts.
[Limits and Background Work] Public API quotas are undocumented. The official Python client maintains 5 requests per second for one token and one endpoint, which can serve as an operational reference but is not a published API limit; this server does not actively rate-limit each call. 429 responses are retried, respecting Retry-After, and the number of retries does not exceed the value allowed by YANGO_RETAIL_MAX_RETRIES. Writes after an uncertain failure are not replayed; network and 5xx retries apply only to side-effect-free reads. After an uncertain write, the current order, price, or inventory should be read first before retrying. Batch limits: products, prices, and discounts are limited to 100 entries per request, and inventory updates to 1,000 rows. There is no background monitoring, and the server only works when the AI application calls it; if the application supports scheduled tasks, order status or inventory can be checked periodically. There is no automatic rollback, and a successful update immediately changes the retailer account. Deletion is limited by the upstream API; products and prices are upserts, and there are no known methods for deleting products, prices, or discounts. Discount support is incomplete upstream, the specific keys for campaign periods and discount values are undocumented, and there are no known methods for listing or deleting discounts; the payload must be confirmed with Yango Tech before use.
[Applicable Scenarios] Suitable for retail operations staff or technical users to query stores, products, prices, and inventory through an AI application, as well as to create orders on demand, update prices and inventory, cancel orders, and read receipts. Technical users can also access a broader range of API methods through raw_request. A Yango Tech Retail account and Bearer token are required,And read-only operations should be performed first; before writing, confirm the target data, and after writing, re-read the current value.
