For developers and AI builders
The Zoopit MCP server connects your AI agent to your own Zoopit data through a single endpoint. Ten curated, read-only tools covering orders, routes, fleet and vehicle positions, instead of a hundred raw API operations for your agent to choose between.
What it is
Den Model Context Protocol is an open standard for giving an AI agent tools. Instead of writing your own HTTP client and tool definitions against the Zoopit API, you point an MCP-capable client at one URL and your agent gets a ready-made tool set with Zoopit vocabulary already built in.
Task-shaped tools
Each tool answers a question a planner would actually ask, rather than mirroring one REST endpoint. “Which routes are running late” is a single call.
Read-only
Nothing the agent does can change an order, a route or a plan. It answers questions; it does not dispatch work.
Your token, your data
The server forwards your own API token and returns only what that token can already see. It grants no access of its own.
Before you start
- A Zoopit account with a project. Every entity in Zoopit belongs to a project, also referred to as
owners, and that is what scopes the agent. - A dedicated API client with read-only scopes. See the warning below before you issue the token.
- An MCP client that supports remote servers with custom headers, for example Claude Code, the Claude Agent SDK, or your own code built on an MCP library.
Create a new API client for the agent. Issuing a token for an API client revokes that client’s previous token. If you mint an MCP token from the client your ERP integration already uses, that integration stops working, and it does so silently. Always create a separate client for the agent.
Connect your agent
Issue a read-only token
In the Zoopit dashboard, create a new API client and give it only read scopes, of the form o/<owner>/read/orders, …/read/routes, …/read/fleet_actors og …/read/gps. The MCP server never writes, so nothing beyond read is useful to it, and a narrow token limits what the agent could ever retrieve.
Point your client at the server
The endpoint is https://mcp.cloud.zoopit.no/mcp, and the token travels as a normal Bearer credential. In a client that reads an mcpServers configuration, that looks like this.
{
"mcpServers": {
"zoopit": {
"type": "http",
"url": "https://mcp.cloud.zoopit.no/mcp",
"headers": { "Authorization": "Bearer YOUR_ZOOPIT_TOKEN" }
}
}
}Keep the token in your secret store or environment, never in a prompt or an agent transcript.
Ask it something
Once connected, the agent can answer directly: which orders are still undelivered today, where is order NL-1417608-LH3, which vehicles are out right now. A request without a valid token is rejected with 401 Unauthorized, which is the quickest way to confirm your token reached the server.
The tools
| Tool | Answers |
|---|---|
search_orders | Which orders are still undelivered? |
get_order | Where is a specific order, and what is on it? |
count_orders | How many deliveries last week, broken down by status? |
search_routes | What ran yesterday? |
get_route | Show me a route’s stops in sequence. |
get_route_progress | Which routes are running late? |
list_fleet | Which vehicles, drivers and schedules do we have? |
get_depots | Which depots do we run from? |
get_vehicle_positions | Where is a given vehicle right now? |
geocode_address | What are the coordinates for this address? |
The server also publishes a glossary as an MCP resource at zoopit://glossary. Zoopit distinguishes an order from a consignment, a task, a booking, an activity and a plannable route, and those words are not interchangeable. Have your agent read the glossary before it interprets results.
What it deliberately does not do
- It does not write. Creating orders, editing routes and triggering optimization runs stay on the API, where a human stays in the loop. See AI Agent Setup for those.
- It is not yet a one-click connector. Authentication is by Bearer token, so it works in clients where you can set a header. Click-to-connect in a hosted chat product needs OAuth, which is not available yet.
- It will not hand over unbounded data. Results are capped per call and rate limited per token, so a looping agent cannot pull your whole order history into a model context.
- It cannot widen your access. The server holds no permissions of its own. An empty result may mean the records do not exist, or that your token cannot see them.
Where your data goes
The token is yours and the agent is yours, so data flows from Zoopit to whichever model provider you have configured. Zoopit is not in that path. Order and consignment records contain recipient names, addresses, email addresses and phone numbers, so treat the connection as an export of personal data and scope the token accordingly.
Getting help
The underlying API is documented at docs.cloud.zoopit.no, and AI Agent Setup covers building directly against it. If you need a tool that does not exist yet, or you are planning something the read-only set cannot answer, get in touch and we will help you scope it.