Getting started with MCP

Updated May 19, 20262 min read

This page gets an AI agent talking to Carriyo in a couple of minutes.

Prerequisites

You need a Carriyo Dashboard login. The agent will act on your behalf, with your permissions, so log in as the user whose access scope you want the agent to inherit.

1. Add the Carriyo MCP server to your agent

The Carriyo MCP server is hosted at:

https://mcp.carriyo.com/mcp

Add it to your MCP-aware agent (Claude Desktop, Claude Code, Cursor, custom frameworks). Most clients use a config file or a UI to register new MCP servers — refer to your client's documentation for the exact mechanics. A typical config block looks like:

{
  "mcpServers": {
    "carriyo": {
      "url": "https://mcp.carriyo.com/mcp"
    }
  }
}

No tokens or secrets in the config — auth is handled interactively in the next step.

2. Authorize the agent

The first time the agent connects, it prompts you to authenticate.

  • If you're not logged into Carriyo — your browser opens the Carriyo login page. Sign in with your Dashboard credentials.
  • If you're already logged in — your browser opens a Carriyo authorization page directly.

Carriyo then asks you to authorize MCP access for the agent. Approve the request to grant access. From this point on, the agent acts on your behalf, inheriting your permissions and tenant / merchant scope.

What the agent can do

Whatever your Dashboard user can do, the agent can do. If you have a read-only role, the agent is read-only. If you can confirm and cancel shipments, so can the agent. See Limits & security for the recommended guardrails.

3. (Optional) restrict the tool catalog

Most MCP clients let you allow / disallow individual tools per server. Use this to lock the agent's surface area down — for example, a diagnostic copilot might only need the read tools, with all write tools disabled.

In Claude Desktop, this is in Settings → MCP servers → Carriyo → Tools. Other clients have similar controls.

4. Try a first call

Once the agent is connected and authorized, ask it something simple:

"Which Carriyo tenants can I access?"

The agent will invoke get_accessible_tenants, which lists every tenant your user can see. This is the canonical first call — the server tells agents to use it as the entry point so any subsequent shipment / report request is scoped to a real, accessible tenant.

A second call to try:

"Show me my most recent five shipments."

The agent will invoke get_shipment_list with a default 30-day window and a small limit, then summarise the response. If you see a list back, you're set.

5. Where to go from here

Production caveat

MCP is best-suited to internal copilots and agentic ops workflows. If you're putting an MCP-driven agent in front of paying customers, build guardrails around the actions an agent can take — see Limits & security for the recommended scope-limiting setup.