> ## Documentation Index
> Fetch the complete documentation index at: https://datum.net/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP server

> Connect Claude Desktop, Cursor, and other MCP-aware AI clients to your Datum Cloud resources with the Datum MCP server.

The [Model Context Protocol](https://modelcontextprotocol.io/) (MCP) lets AI
clients like Claude Desktop and Cursor talk to your Datum Cloud resources
directly, outside the terminal. Datum publishes an official **Datum MCP
server** for this — a standalone binary you point your MCP-aware client at.

Where the [AI assistant](/datumctl/ai/assistant) is the interactive,
terminal-first experience built into `datumctl`, the Datum MCP server is the
bridge that gives a desktop AI assistant secure, standards-based access to your
organizations, projects, domains, proxies, DNS, and more.

<Info>
  The Datum MCP server is distributed as a separate `datum-mcp` binary, not as a
  `datumctl` subcommand. Installation, the full tool catalog, and one-click setup
  for clients such as Cursor live on the dedicated [Datum MCP](/datum-mcp) page
  and upstream in the [Datum MCP repository](https://github.com/datum-cloud/datum-mcp).
</Info>

## When to use the MCP server

Reach for the MCP server when you want an AI client that lives outside your
terminal to work with Datum Cloud:

* **Claude Desktop** — chat with an assistant that can read and change your
  Datum resources.
* **Cursor** and other IDE-embedded assistants — bring Datum context into your
  editor while you build.
* Any other MCP-aware client that speaks the standard stdio transport.

If you would rather stay in the terminal, `datumctl` has a built-in assistant
that answers questions in natural language without a separate MCP client — see
the [AI assistant](/datumctl/ai/assistant) guide.

## What it exposes

The Datum MCP server gives connected clients tools to inspect and manage your
Datum Cloud resources. Clients can understand the current state of your
resources, suggest and implement changes, and help diagnose common
configuration issues. The catalog currently covers:

* Organizations and organization memberships
* Users
* Projects
* Domains
* HTTP proxies, HTTP routes, and gateways
* Traffic protection policies
* DNS zones, DNS record sets, and DNS zone classes
* APIs (list and describe available resource types)

<Note>
  The tool catalog evolves with the server. See the [Datum MCP](/datum-mcp)
  page for the authoritative, up-to-date list and the exact operations each tool
  supports.
</Note>

## Connecting a client

Most MCP clients launch the server over stdio. Add an entry like this to your
client's MCP configuration:

```json theme={null}
{
  "datum-mcp": {
    "command": "datum-mcp",
    "args": []
  }
}
```

<Steps>
  <Step title="Install the Datum MCP server">
    Follow the install instructions on the [Datum MCP](/datum-mcp) page. Binaries
    are available for Linux, macOS, and Windows on x86 and arm64.
  </Step>

  <Step title="Add the server to your client">
    Point your MCP client at the `datum-mcp` binary using a stdio configuration
    like the one above. Cursor users can use the one-click install link on the
    [Datum MCP](/datum-mcp) page.
  </Step>

  <Step title="Sign in on first use">
    The first tool call triggers a browser-based sign-in (see below). Approve it,
    and your client is connected.
  </Step>
</Steps>

## Authentication

On first use, the server runs a browser-based OAuth (PKCE) sign-in, then stores
credentials — including a refresh token — securely in the system keychain.
Subsequent calls reuse and refresh the token automatically, so you sign in once.
This is the same security model `datumctl` uses when you
[log in](/datumctl/auth/logging-in): no static API keys.

The server implements the latest
[MCP Authorization](https://modelcontextprotocol.io/specification/2025-06-18/basic/authorization)
specification and supports OAuth 2.1 (PKCE).

<Warning>
  The MCP server acts on real Datum Cloud resources with your credentials.
  Connect it only to AI clients you trust, and review any changes an assistant
  proposes before approving them.
</Warning>

## Next steps

* [Datum MCP](/datum-mcp) — set up the Datum MCP server: install, one-click client setup,
  and the full tool catalog.
* [AI assistant](/datumctl/ai/assistant) — prefer the terminal? Use the built-in `datumctl ai`
  assistant instead.
* [Logging in](/datumctl/auth/logging-in) — review how Datum Cloud handles credentials before
  your first connection.