Skip to main content
The tools are the actions your AI agent can take in Zippendo for you, such as creating a shipment or reading an order. The MCP server offers one tool for each action in the API, and your agent only sees the tools that the permissions you approve allow.

​
How tools map to permissions

Every tool needs a permission (a scope) that matches the resource and the action:
  • Reading data needs a read: scope. Listing shipments, for example, needs read:shipments.
  • Changing data needs the matching write: scope. Creating a shipment, for example, needs write:shipments.
Scopes build on each other, so you can grant broadly or narrowly:
  • write:all includes read:all and every write: scope in the table below.
  • read:all includes every read: scope in the table below.
  • write:<resource> includes read:<resource> (write:shipments includes read:shipments).
An agent granted only read:shipments sees the shipment read tools (list, details, tracking, documents) but none of the tools that create, send or delete shipments.

​
Scope reference

The consent screen can also list other permissions, such as organizations, team members (org_members), return portals and API tokens. They unlock no MCP tools today.
Grant the narrowest set that does the job. read:all plus write:shipments is a good default for an assistant that reports on everything but only acts on shipments.

​
Example prompts

Once connected, talk to your agent in plain language. It picks the right tools and asks Zippendo:
  • “How many shipments are still pending, and which carriers are they on?”
  • “Create a shipment for order #10421 with PostNord and give me the tracking number.”
  • “Split the shipment for order #10388 into two parcels and send both.”
  • “Which orders from the last 24 hours don’t have a shipment yet?”
  • “Track shipment shp_8f3kd9 and tell me if it’s delayed.”
  • “Add a shipping rule that sends orders over 10 kg with GLS.”

​
Documentation and SDK tools

Besides the action tools, the MCP server lets an agent read Zippendo’s documentation, so it can help you build an integration in your programming language. These tools need no scope and every connected agent has them. The agent is told to prefer the SDK for your language, and to use the raw OpenAPI spec only when there is no SDK for it. The content comes live from zippendo.com/docs and the SDK repositories, so it’s always current. Try prompts like:
  • “Using the Zippendo Python SDK, write a script that creates a shipment for an order and prints the tracking number.”
  • “How do I authenticate to the Zippendo API in Go? Show me the SDK setup.”
  • “I’m on PHP. Write a webhook handler that verifies the signature and reacts to shipment.dispatched.”

​
Safety model

Acts as you

Tool calls run as you, only with the permissions you approved.

One organization

A connection belongs to one organization. It can’t read or change another organization’s data.

No staff access

An agent can never perform Zippendo staff actions.

Full validation

Every call runs the same sign-in, plan limit and validation checks as the app.
Deleting shipments, orders, carriers, addresses, shipping rules or roles, sending a shipment, and creating or revoking checkout tokens and signing secrets all need an explicit confirmation in the tool call, so the agent should ask you first. Some sensitive actions, such as billing and API token management, are left out of the tools entirely and no agent can use them, whatever its permissions.

​
The same tools power Zippy

The MCP server shares its tools with Zippy, the assistant in the Zippendo app. Only the place differs: Zippy runs inside the app, while MCP lets your own AI tool use the same actions. Both apply the same permissions, organization limits and validation.

Connect your agent

Ready to set this up? Follow the setup guide for your tool.