Skip to main content
The Sendcloud MCP server exposes the Public API v3 to AI assistants and agents through the Model Context Protocol (MCP). Every API operation (for example creating shipments, listing orders, tracking parcels, and more) is available as a tool your MCP client can call, so an agent can act on your Sendcloud account in natural language without you writing any HTTP code. It is an aggregated server: a single endpoint that bundles every Public API v3 domain as a separate toolset. You connect once and get access to all of them, or narrow the connection to just the toolsets you need.
This page describes how to use the MCP server as an agent or MCP client. For the underlying endpoints and schemas, see the API v3 reference.

​
Endpoint

The server speaks the standard MCP protocol over HTTP, so most MCP clients can connect to it, including Claude, Claude Code, ChatGPT, VS Code, and agents built on an MCP SDK.
Your MCP client must support OAuth Client ID Metadata Documents (CIMD) to sign in. Some clients don’t support CIMD yet and can’t connect to the Sendcloud MCP server, for example Cursor.
Before you connect – please read.Results are reference information only. This connector lets your AI assistant read data already available to you via the Sendcloud API (e.g. shipment status, tracking, rates, order details). That data reflects a snapshot at the time your AI assistant made the request and can change afterwards – for example, carrier surcharges, tracking status, and delivery estimates update over time. Do not treat a response from your AI assistant as a confirmed or binding figure; for anything time-sensitive or in dispute (e.g. a surcharge query), check your Sendcloud account directly or contact Sendcloud support.Sendcloud does not control your AI assistant’s output. We cannot control, and are not responsible for, how your chosen AI assistant interprets, summarizes, or responds to this data, including any follow-up questions or attempts to get it to “confirm” a figure.You choose the AI tool. You are solely responsible for selecting the AI assistant you connect to this feature. Sendcloud does not select, endorse, or control that AI provider, and has no visibility into or control over how it subsequently stores, uses, or discloses the data it receives – that is governed by your own agreement with that provider, not by Sendcloud.By connecting, you confirm you have read and accept the above.

​
Connecting

1

Add the server to your MCP client

Point your client at the MCP URL. For example, with the Claude Code CLI:
Other clients (Claude Desktop, custom agents) accept the same URL in their MCP server configuration. A typical JSON config looks like:
2

Authorize with OAuth 2.0

The first time your client connects, it discovers the authorization server automatically (via RFC 9728 protected-resource metadata) and walks you through an OAuth 2.0 sign-in at account.sendcloud.com. The consent screen lets you pick which permission groups to grant (see Choosing permissions). After you approve access, the client stores the resulting token and attaches it to every tool call.Your credentials live in the MCP client, never in the server. See Authentication for how Sendcloud OAuth 2.0 works.
3

Discover and call tools

Once connected, the client lists the available tools. Ask the agent to perform a task (for example “list my most recent orders” or “track parcel 123456789”) and it selects and calls the matching tool, forwarding the request to the Public API v3 and returning the response.

​
Selecting toolsets

By default the server exposes all toolsets. To limit a connection to a subset (for example keeping the agent’s tool list focused and relevant) add a ?toolsets= query parameter with a comma-separated list of toolset names:
The filter applies to both tool discovery and tool calls: a tool outside the selected toolsets is hidden from the list and rejected if called directly.

​
Choosing permissions

When you authorize the connection, the consent screen lists permission groups such as Orders, Shipping, and Returns. Each group grants a set of OAuth 2.0 scopes, and the token your client receives can only do what those scopes allow. Toolsets control which tools the agent sees. Scopes control what the agent is allowed to do with your account. For example, you can let an agent read your parcels without allowing it to create or announce shipments.
Grant only the groups your agent needs. A tool that needs a scope you didn’t grant fails with a permission error, which is the expected result.
To change your permissions, re-authenticate the connection in your MCP client and select a different set of groups. See Scopes for OAuth apps for the scopes in each group.

​
Available toolsets

Each toolset maps to one Public API v3 domain. Its tools need the scopes for that domain, as listed in Scopes for OAuth apps.

​
How tools map to the API

Each tool corresponds to one HTTP operation. The tool name is derived from the operation’s operationId, and is namespaced by toolset to avoid collisions — for example orders_sc_public_v3_orders_get_list_orders (list orders) or shipments_sc_public_v3_scp_get_shipment_by_id (retrieve a shipment).The operation’s path, query, and header parameters — plus the request-body fields — become the tool’s arguments. Required API parameters stay required on the tool.
Tools are annotated with hints about their effect, derived from the HTTP method: read-only (GET), idempotent, or destructive (DELETE, POST). MCP clients use these to auto-confirm safe lookups while prompting you before a tool that creates, cancels, or deletes data.
Tools that create or cancel shipments act on your live account. As with the API directly, you may be invoiced for shipments you create if they aren’t canceled within the cancellation deadline.

​
Next steps

Authentication

How Sendcloud OAuth 2.0 works

Scopes

The permissions each consent group grants

API v3 reference

The endpoints behind every MCP tool