AI Agents
The agent connects through our MCP server as a signed-in user or as a standalone agent with its own permissions and token, and works on the same orders and rules as your team.
Capabilities
Carriyo lets AI agents do the work of order and delivery operations: chasing late shipments, clearing the review queue, moving fulfillment orders between stores, and booking and reassigning carriers. It is for operations and technology leaders who want agents on the routine work without handing them more reach than a new employee would get.
Our MCP server exposes 90+ tools that cover lookups, delivery options, products, inventory, orders, fulfillment orders, returns, shipments, reports, automation rules and system logs.
Read an order and its fulfillment orders, hold and release orders.
Move a fulfillment order to another location, split or merge fulfillment orders.
Create and confirm shipments, reassign or reprocess a shipment, update the promise date a customer sees.
Handle returns and edit checkout delivery options.
Run reports such as SLA compliance and carrier performance.
Automation rules and hold rules are readable by the agent and editable only in the dashboard.
Permissions
An agent connects to our MCP server in one of two ways, and in both it acts only inside the permissions it was given. Every MCP call is logged like a dashboard action, tagged with the user or the agent that made it, and visible in the Integration Monitor.
Signs in through OAuth 2.1 with PKCE as a real dashboard user and inherits that user's role, tenant and merchant scope. For agents a person runs from a desktop.
Created in Carriyo with its own permission set and an access token you can rotate at any time. For agents running from your own cloud, with nobody signing in.
Late shipments
An agent lists shipments and reads their late flags, pulls each one's activity timeline and the carrier logs behind it, and works out why each is late. Then it can reprocess a shipment in error, reassign an unshipped one to another carrier account, schedule a new delivery date with the carrier, and write a short brief for the person who owns the queue.
What it reads is the same for every carrier, because Carriyo normalises carrier events into one set of statuses across 140+ carriers, and lateness has one definition.
We would not have an agent move a customer's promised date on a doubtful carrier scan; better to keep the promise, query the carrier and act on something reliable.
Order holds
Hold rules catch the predictable cases as orders are created and hold them with an assignee, a priority and a review deadline. While an order is held, every update and every fulfillment operation is rejected until it is released or cancelled, and that applies to the agent as much as to anyone else.
An agent has two jobs here: it prepares the queue so the person spends their time on the decision, and it raises its hand when it sees something the rules did not anticipate.
Whether the agent may release is your choice: leave the release tool off and every release is made by a person, or switch it on for the clear cases once the audit trail earns it.
Storefront API
A buying agent choosing where to place an order needs delivery choices it can read, and our Storefront API returns exactly that: the options for this basket to this address, each with a price, an estimated delivery date and, for collection, the nearest pickup points. The promise it reads is the one our order management keeps.
Agent-ready stack
Four things have to be true before you would let an agent near a live order. The lower-risk way to start is to put rules and holds in place and start an agent on read-only work such as briefs and queue preparation, then widen what it may do as the audit trail earns it. This layer has been shaped by the orders and shipments of 100+ brands in 200 countries, and it does not change depending on who, or what, is operating it.
The advantage is not in the agent, which everyone can buy, but in what it stands on.
YOUR WAY. RULES. OR AI.
The pattern we design for runs all three modes on the same orders: the rules make the routine decision, the agent reads the exceptions and prepares them, and a person decides what the rules did not foresee.
Connect Claude Desktop, Claude Code, Cursor or your own agent to our MCP server, as a Carriyo user through OAuth 2.1 with PKCE or with a standalone agent's token.
A signed-in agent takes that user's role, tenant and merchant scope; a standalone agent has the permission set you gave it. Both get only the tools you switch on.
It reads orders, fulfillment orders, shipments and carrier logs, prepares the queue, and acts within the holds, rules and permissions your team already runs on.
Every MCP call is logged like a dashboard action, tagged with the user or the agent that made it, and visible in the Integration Monitor.
A person does the work by hand and sees everything.
The business writes its decisions down once and the platform applies them to every order without anyone looking.
Through our native MCP layer, an agent reads the same data, follows the same rules and takes the same actions a person would, within the permissions it has been given.
30%
fewer customer service calls at Sephora Middle East
Proof point
Sephora Middle East runs 5 fulfillment centres, 92 stores and 120k+ orders a month on Carriyo across 10+ carriers, with 30% fewer customer service calls. The same MCP layer reaches every order and shipment on the platform.
AI Agents
Read orders and fulfillment orders, hold and release orders, move, split or merge fulfillment orders, create, reassign and reprocess shipments, update promise dates, handle returns and run reports, all within the permissions the agent was given.
No. Connect Claude Desktop, Claude Code, Cursor or any MCP-compatible client and sign in as a Carriyo user. Your own agent connects the same way, as a signed-in user or as a standalone agent with its own token.
No. A signed-in agent inherits its user's role, tenant and merchant scope; a standalone agent has only the permission set you created it with. Give either the smallest set it needs and switch off tools it should not have.
Yes. Create a standalone agent in Carriyo with its own permission set and a dedicated access token, and connect it from your own cloud or orchestration layer. The token can be rotated at any time, and every call it makes is logged under the agent's own identity.
Order holds stop an order before anything is picked and reject every update until it is released, and release can stay in human hands. Every agent call is logged like a dashboard action.
It is the operations side of it. Buying agents choose merchants they can read, and our Storefront API gives them structured delivery options and promises. Operating agents run the work after the order through our MCP layer, with the same tools, rules and permissions as your team.
To go through which of your post-purchase jobs an agent could safely take on, and where a person should stay in the loop, get in touch.