LLM-Native Operations in Merchant Portal

MCP (Model Control Plane) for Merchant Portal enables merchants to operate Brink entirely through their own Large Language Model (LLM) over HTTP, without relying on a graphical user interface.

Through MCP, an LLM becomes a first-class operational client to Brink. It can securely read data, query insights, and execute real operations by interacting with Brink’s Management and OMS APIs using natural language translated into validated API calls.

This allows merchants to run their daily commerce operations via conversation, scripts, or AI agents—searching orders, managing deliveries and refunds, configuring markets, inventory, pricing, and campaigns—without clicking through UI workflows.

MCP is not a chatbot layer. It is a controlled execution plane that exposes Brink’s domain logic, schemas, and permissions to LLMs in a safe, auditable, and deterministic way, governed by user roles and access.

Core Capabilities

• Data access via language

• Search and filter orders

• Retrieve customers, products, prices, inventory, campaigns

• Query operational and performance insights

• Operational execution

• Create shipments and deliveries

• Perform refunds and cancellations

• Transition OMS states

• Configure inventory rules and availability

• Set up new markets, currencies, store groups, and channels

• Create or modify campaigns and discounts

• LLM-native API interaction

• Natural language → validated API requests

• Schema-aware payload generation

• Permission-scoped execution

• Full auditability of actions

• HTTP-based integration

• Works with merchant-owned LLMs

• Supports AI agents, scripts, automations, and internal tools

• No dependency on Merchant Portal UI

Spotlight: Discount Operations

Discount management is the single highest-friction area in day-to-day Brink operations. The majority of merchant support requests fall into one of four patterns: configuring rules, simulating outcomes, diagnosing unexpected behavior, or reporting on usage. MCP collapses each of those into a single LLM prompt — backed by the same permission model and audit trail as direct API access.

Configuration

• "Create code KLARA10K, capped at 10,000 SEK, calculated on regular price even during active campaigns, valid only on these SKUs."

• "Set up a Gift-with-Purchase rule that adds the size-matched mystery item when cart total exceeds threshold X."

• "Add an OR condition to discount rule Y so it triggers for either VIP or member tier."

Simulation

• "If a customer with the VIP tag adds 4 pairs of socks in size 36–40, which mystery variant is selected and what is the final cart total?"

• "Show how stacked codes A and B combine on a 1,200 SEK cart with express shipping in the SE market."

• "Preview the discount outcome for this exact cart before I commit the rule change."

Diagnostics

• "Why has code X been used more times than its usage limit?"

• "List all reasons code Y was rejected at checkout in the last 24 hours."

• "Why is the express-shipping member discount not applying for this customer on the FI market?"

• "Find all discount rules where the cache status is stale."

Reporting

• "Top 10 discount codes last month by revenue, split by store market."

• "Usage and revenue trend for cart rule X over the last 90 days."

• "Which discount rules are active right now in production, and which ones are scheduled for the next 7 days?"

This replaces what is today multi-step CX work — switching between the rule editor, order list, sales dashboard, and ad-hoc API calls — with a single language interface that preserves auditability end-to-end.

Typical Use Cases

• “Find all failed orders from yesterday and explain why they failed.”

• “Refund this order and notify the customer.”

• “Create a new market for Canada with CAD pricing and default shipping.”

• “List all orders delayed due to inventory constraints.”

• “Adjust safety stock rules for this warehouse.”

• “Generate a campaign payload for 20% off selected SKUs.”

• “Find every order where discount code SUMMER25 was applied with the wrong base price and list the impact in revenue.”

Benefits

Operational speed

• Execute complex workflows in seconds instead of navigating multiple UI views

• Faster issue resolution and daily operations

Lower operational friction

• Work directly in language, scripts, or AI agents

• No UI learning curve for advanced users

Automation-ready by design

• Same interface supports human-driven commands and automated agents

• Natural path from assisted actions to full automation

Merchant control & safety

• Scoped permissions and explicit execution boundaries

• Explainable actions with full traceability

Future-proof operations

• Enables AI-first operating models

• Allows merchants to build their own operational assistants on top of Brink

Positioning

MCP for Merchant Portal turns Brink into an LLM-native commerce operations platform—where managing orders, inventory, markets, and insights can happen entirely through language, APIs, and automation, without a traditional UI.