mcp-yango-delivery

Yango Delivery MCP (A1 Yango Delivery MCP) is an MCP server used to connect AI applications to Yango Delivery enterprise accounts. Users can estimate delivery prices, book couriers, track routes, and share tracking links with recipients using natural language. This server is only applicable to the international Express Claims API and does not support next-day delivery, pickup points, or parcel lockers.

Key capabilities

• A total of 13 tools: 7 covering the delivery lifecycle, 5 for tracking and contacting couriers, and 1 tool for direct API requests, raw_request.
• Pre-order estimates: check current prices, route distance, and estimated time of arrival (ETA) without creating a delivery claim.
• Separation of creation and ordering: prepare and review a claim, then start searching for a courier.
• Track active deliveries: get status, courier location, ETA for each route point, temporary phone number, confirmation code, and recipient tracking link.
• Check and cancel deliveries: determine whether cancellation is free, charged, or no longer available, and cancel a claim using the current version and the cancellation mode returned by the API.
• raw_request covers methods without dedicated tools, including rates, delivery methods, proof of delivery, claim editing, and return operations; it can be used to create or change real data, and is intended for technical users who understand the upstream API.

How to use Requires Node.js 20+, a Yango Delivery enterprise account, and an integration token. After obtaining the integration token, add the server to the AI application. The common startup command is npx -y mcp-yango-delivery@latest, with the environment variable YANGO_DELIVERY_TOKEN, using STDIO. The repository provides configuration instructions for various AI clients. You can start with example questions, such as: Estimate delivery for a 2 kg parcel from [pickup address] to [delivery address].

Delivery process

1. Estimate: a price check returns the current quote, distance, and ETA without creating anything.
2. Create: the claim is added to the account and enters the estimation process; by default, no courier has been ordered yet.
3. Review: read the latest claim status, price, and version; a successful claim enters an approvable state.
4. Accept: after acceptance, the search for a courier begins; from then on the delivery has been ordered and actual charges may be incurred.
5. Track: after a courier is assigned, the server can return location, ETA, contact details, and the recipient linkContinued.

Limitations and Notes

• Pricing quotes typically expire about ten minutes after estimation. If auto_accept: true is passed when creating a claim, the separate review and acceptance steps are skipped, and a courier may be dispatched automatically.
• The server provides MCP annotations for read, write, and destructive operations; the AI client decides when and how to request confirmation.
• Claim creation uses a unique request_id; reusing the same ID returns the same claim. Acceptance and cancellation are not automatically retried after network errors or 5xx responses, because the operation may have already succeeded; the current claim should be read before retrying.
• The integration token is stored in plaintext in the AI client's local configuration and should be treated as a password; never commit a configuration containing a real token.
• The server runs on the user's machine and sends claim data directly to b2b.taxi.yandex.net; the integration token is attached only to requests sent to the configured API host. raw_request accepts relative paths and blocks requests that resolve to other hosts.
• By default, anonymous technical telemetry is sent 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. Integration tokens, claim data, tool parameters, and prompts are not read or sent. Telemetry times out after two seconds and does not block tool calls. It can be disabled via configuration.
• There is no documented sandbox environment; claim creation and acceptance should be treated as production operations. Quotes are time-limited, and the current claim should be read before accepting.
• Tracking requires an active claim; before courier assignment or after claim completion, location, ETA, contact details, and tracking links may be unavailable.
• No numeric API quotas are published. When the API returns 429, the server follows Retry-After (if present) and retries according to YANGO_DELIVERY_MAX_RETRIES.
• No background monitoring: the server works only when the AI application calls it. If the application supports scheduled tasks, active claims can be checked periodically.
• No automatic rollback: after an uncertain acceptance or cancellation result, the claim should be read first before repeating the operation.

Technical Information The project provides the complete schema, response fields, and status details (see the tool reference). The integration token is bound to the business account and does not expire by itself, but changing the account password resets it.