Claude Code

Claude Code is Anthropic's CLI for AI-assisted software engineering. It supports MCP servers at three scopes: local (your machine), project (shared through .mcp.json in a repo), and user (your global ~/.claude.json).

Prerequisites

• Claude Code installed. For more information, see claude.ai/code.
• The ionoscloud-mcp binary built and available. For more information, see Quick Start.
• An IONOS CLOUD API token. For least-privilege access, see Authentication to scope the token to only the tools you need.

Add the MCP server

Option A: CLI (recommended)

Use the claude mcp add command to register the server:

claude mcp add ionoscloud /usr/local/bin/ionoscloud-mcp \
  --env IONOS_TOKEN=your-api-token

To also enable Object Storage tools:

claude mcp add ionoscloud /usr/local/bin/ionoscloud-mcp \
  --env IONOS_TOKEN=your-api-token \
  --env IONOS_S3_ACCESS_KEY=your-access-key \
  --env IONOS_S3_SECRET_KEY=your-secret-key

By default, this registers the server at local scope (your machine only). To share the config with a team via source control, use --scope project instead; this writes to a .mcp.json file in the current directory.

Option B: Project config file

Create or edit .mcp.json in your repository root:

{
  "mcpServers": {
    "ionoscloud": {
      "command": "/usr/local/bin/ionoscloud-mcp",
      "env": {
        "IONOS_TOKEN": "your-api-token"
      }
    }
  }
}

Commit this file to share the server configuration with your team. Each team member supplies credentials via their own environment variables; never commit tokens to the file. Use ${env:IONOS_TOKEN} in the config to reference an environment variable instead of hardcoding the value:

{
  "mcpServers": {
    "ionoscloud": {
      "command": "/usr/local/bin/ionoscloud-mcp",
      "env": {
        "IONOS_TOKEN": "${env:IONOS_TOKEN}"
      }
    }
  }
}

Each team member then exports IONOS_TOKEN in their shell profile (~/.zshrc, ~/.bashrc) and the token is never stored in the committed file.

Note: Claude Code prompts you to approve a project-scoped .mcp.json on first use. When the prompt appears, review the command path and env block; confirm the binary path is correct and no unexpected variables are present. Select Allow to proceed or Deny to cancel (the server will not load if denied). For CI/CD environments, use the --allowedTools flag or a pre-approved trust policy instead of interactive approval.

Verify the connection

1

List available MCP servers

claude mcp list

ionoscloud appears in the output when the server is registered.

2

Start Claude Code and test

claude

Then type: List my IONOS CLOUD data centers.

Result: Claude Code calls list_datacenters and returns your data centers.

Note: All tools are registered at startup by default, so Compute Engine and Object Storage tools appear immediately. To reduce the startup tool count, pass --env IONOS_MCP_LOAD_MODE=lazy when adding the server (or add it to .mcp.json) and call the ionos_load_compute_tools or ionos_load_objectstorage_tools sentinels on demand. For more information, see Selective Tool Loading.

Deploying for a team

The project-scoped .mcp.json file is the recommended way to share the MCP server configuration across a team. Because the server is read-only, sharing access to production account data carries no risk of accidental modifications.

Recommended setup:

1. Add .mcp.json to the repository root using ${env:IONOS_TOKEN} interpolation so no credentials are committed:

   {
     "mcpServers": {
       "ionoscloud": {
         "command": "/usr/local/bin/ionoscloud-mcp",
         "env": {
           "IONOS_TOKEN": "${env:IONOS_TOKEN}"
         }
       }
     }
   }

2. Each team member exports their own token in their shell profile:

   export IONOS_TOKEN="their-own-api-token"

3. Commit .mcp.json to the repository. Claude Code prompts each team member to approve it on first use; they review the config and select Allow.

For CI/CD pipelines: Pass the token as an environment variable in your pipeline and use claude --allowedTools ionoscloud to pre-approve the server non-interactively.

For full Claude Code MCP documentation including transport options and dynamic tool updates, see Connect Claude Code to tools via MCP.