How It Works - Clawdot虾点·文档中心

End-to-End Workflow

Super Agent connects users, AI, and the merchant network through Clawdot Gateway to complete a full transaction workflow:

All amount fields are in cents (integer). For example, payable_price: 990 means ¥9.90.

Clawdot’s Abstraction Layer

When AI Agents directly connect to merchant platforms, complexity arises. Clawdot Gateway provides a unified abstraction in the middle, making Agent work simple:

What Agent Sees	What Gateway Handles
Simple `shop_id` + `cart_id`	Multiple ID format conversion (encrypted ID, internal ID, third-party ID); coordinates packed into `cart_id`
Standardized menu structure	Specs/attrs/ingredients parsing, mutual exclusion rule calculation
`ingredient_option_ids` one-tap ordering	Traverse all spec groups, exclude mutually exclusive items, select defaults
One final price	Two renders: first get coupon list, then re-render with optimal coupon
Unified authentication method	API Key and consent grant validation, request signing, User-Agent adaptation
Unified error codes	Categorize various platform errors into standard error codes

Super Agent’s Role

MCP Access

Clawdot Gateway exposes its capabilities over MCP (Model Context Protocol) — the AI-native calling method designed for AI models like Claude. Super Agent connects with an MCP client (Streamable HTTP) to the public endpoint https://eleme-gateway.hicaspian.com/mcp/v1. Each request carries Authorization: Bearer clw_... in the HTTP header to identify the Agent, while user authorization is passed as the consent_grant_id argument on each business tool:

# Merchant search / browse
search_shops(consent_grant_id, keyword, address_id, lat, lng)

# Shop menu
get_shop_menu(consent_grant_id, shop_id, cart_id)

# Address management
search_addresses(consent_grant_id, keyword, lat, lng)   # returns saved_addresses + suggestions[].token
select_address(consent_grant_id, suggestion_token, contact_name, contact_phone)

# Quote and order
quote_cart(consent_grant_id, shop_id, cart_id, address_id, items)
preview_order(consent_grant_id, shop_id, cart_id, address_id, items)
create_order(consent_grant_id, preview_id, confirmation_token)
get_order_status(consent_grant_id, order_id)

Tool parameters and return values are optimized for how AI models call them. For the full tool list, see MCP Overview.

MCP tools run on the same underlying gateway service layer, exposed externally only through the MCP interface.

Workflow Details

The full ordering flow is a stateful chain — each step’s returned ID must be passed verbatim to the next: search_addresses → select_address → search_shops → get_shop_menu → quote_cart → preview_order → create_order. Fix the address first, then search deliverable shops by that address, so downstream tools all have the delivery coordinates. For full field tables, see Order Flow.

Address Management Phase

Fix “where to deliver” first; downstream shop search and quote / preview all rely on this address. First use:

Call search_addresses to search the address the user mentioned; it returns suggestions[].token
User selects from the suggestion list
Call select_address (with suggestion_token + contact) to register the delivery address, which returns an address_id

Subsequent uses: search_addresses also returns saved_addresses[] (plus nearest_address_id when lat/lng are given), so you can reuse an existing address_id and reduce interaction steps.

Search Phase

Agent calls search_shops (with the address_id from the previous step) to search deliverable merchants by the delivery address. Without keyword it browses (returns up to 20 nearby shops with decision fields like distance, rating, delivery fee); with keyword it does a precise search (~5 shops, accepts a shop name, category, or concrete item name). Each result carries a shop_id and a cart_id; the cart_id packs the shop and delivery coordinates and is the entry point for later steps.

Menu Browsing Phase

Quote + Preview + Order Phase

This is the core stateful chain:Quote → quote_cart

Input: shop_id + cart_id + address_id + items (list)
Output: quote_id + price details

Preview → preview_order

Input: shop_id + cart_id + address_id + items (optionally pass quote_id to validate the quote context)
Output: final price (in cents) + preview_id + confirmation_token (must be used together, valid for about 10 minutes)

Order → create_order

Input: the preview_id + confirmation_token from the same preview
Output: order_id, status, and (when payment is needed) payment_action

You must display the price to the user and get explicit confirmation before calling create_order. The confirmation_token is the idempotency key — one token can place only one order.

Order Tracking Phase

After placing an order, Agent can call get_order_status to query order status:

Waiting for merchant to accept
Merchant is preparing order
Rider has picked up food
Rider is delivering
Delivered

Can be used to proactively update users on order progress. You can also pass callback_url to create_order, and every status change on the order will POST a callback.

Passwordless Signing Phase

Passwordless signing is an account-level, one-time prerequisite: sign once, and later passwordless-payment orders can be charged automatically — it is not bound to any single order and is not done per order, so it stands outside the ordering chain above (do not treat it as the last step of placing an order). Check with get_sign_status whether already signed first, and only if not, call get_sign_action (with return_url, no order_id) to get a signing action. When action_type is open_h5, direct the user to action.action_url to complete signing; the browser then returns to return_url. When it is none, signing is already done. Then poll the result with get_sign_status.

Create an Agent in the Portal to get an API Key; obtain a user consent grant via the binding flow (SMS code or H5). For the full chain and fields, see Order Flow and Authentication.

​End-to-End Workflow

​Clawdot’s Abstraction Layer

​Super Agent’s Role

​MCP Access

​Workflow Details

End-to-End Workflow

Clawdot’s Abstraction Layer

Super Agent’s Role

MCP Access

Workflow Details