IONOS CLOUD MCP Server

What Is It?

IONOS CLOUD MCP Server is a local binary that implements the Model Context Protocol (MCP), connecting AI clients and autonomous agents to a customer's IONOS CLOUD account via natural-language prompts or programmatic agentic loops. It exposes 110 read-only tools across 6 IONOS CLOUD products and translates AI tool calls into direct HTTPS requests to the IONOS CLOUD API - no third-party service sits in the data path. The binary runs entirely on the user's workstation (macOS, Linux, or Windows) and communicates with the AI client over stdio using the MCP JSON-RPC protocol. Source code is publicly available under the Apache-2.0 license.

Quick Facts

Tool Coverage by Product

Selective Tool Loading Modes

Always-on tools in lazy mode (37): Cloud DNS, Billing, Certificate Manager, and Activity Log tools, plus the two loader sentinels (ionos_load_compute_tools, ionos_load_objectstorage_tools).

What You Can Do

Query Live Infrastructure via Natural Language

Ask an AI client questions such as "List all servers in my data center" or "Show certificates expiring in 60 days." The binary translates each tool call into an IONOS CLOUD API request and returns structured JSON for the model to reason over. All 110 tools are read-only, so agents can safely audit production environments without risk of accidental changes.

Build Agentic and CI/CD Workflows

Embed the MCP server in automated pipelines. The binary runs headlessly as a subprocess from any MCP client SDK (Python, TypeScript, etc.), enabling build scripts to fetch current resource state, generate configuration diffs, or produce deployment manifests based on live data. Use IONOS_MCP_LOAD_MODE=lazy to keep the tool surface minimal in unattended pipelines.

FinOps Billing Analysis (FOCUS v1.3)

Pull raw billing data through the Billing tools and load the FOCUS v1.3 (FinOps Open Cost and Usage Specification) column-mapping file via the MCP resource URI ionos://billing/focus-v1.3. The assistant can transform the data into the exact CSV or JSON format required by a FinOps platform in a single automated loop.

Combine Live Data with Documentation

Pair the infrastructure server with the companion Documentation MCP server (no auth required, hosted at https://docs.ionos.com/cloud/~gitbook/mcp) to give the AI simultaneous access to live resource data and the full IONOS CLOUD product knowledge base. A single prompt can retrieve the recommended token scopes from the docs and then immediately call live tools to validate the current setup.

Sovereign AI Workflows

Keep all API traffic inside IONOS CLOUD's EU infrastructure by pairing the local binary with IONOS CLOUD AI Model Hub. The binary calls only the IONOS CLOUD API over HTTPS (EU-resident endpoints). Pair it with an AI client that supports a custom OpenAI-compatible inference endpoint (such as Continue or OpenCode) and point inference at the AI Model Hub to ensure neither API calls nor model prompts leave EU jurisdiction.

Activity Log Access

Query recent account events via the Activity Log tools. The default window is 7 days and the default event count per call is 25. Activity Log access requires the ACCESS_ACTIVITY_LOG IAM privilege on the token.

Best For

Consider Alternatives If

Key Considerations

Billing & Costs

• Main billing: No separate charge for the MCP server binary itself (Apache-2.0, free to use)
• Downstream costs: All IONOS CLOUD resource usage queried through the server is billed under the respective wrapped products (Compute Engine, Object Storage, etc.) per their standard rates
• Additional note: The binary does not appear in the IONOS CLOUD price list as a standalone line item. Verify with official documentation if this changes.

Limitations

• Read-only: All 110 tools are read-only. Write operations (create, update, delete) require the direct IONOS CLOUD API, SDKs, or the DCD console.
• Tool cap on some clients: The default eager mode registers 110 tools. Clients with a hard 100-tool limit (such as Windsurf) will reject startup; set IONOS_MCP_LOAD_MODE=lazy (39 tools at startup) before use.
• Lazy mode caveat: In lazy mode, an agent that has not called the appropriate loader sentinel will report "I don't have a tool to..." rather than an error. Explicitly prompt the assistant to load Compute or Object Storage tools before those operations are needed.
• Internet required: The binary must be able to reach the IONOS CLOUD API over outbound HTTPS. Air-gapped workstations are not supported.
• Activity Log privilege: Activity Log tools require the ACCESS_ACTIVITY_LOG IAM privilege on the API token. Tokens without this privilege will fail those specific calls.
• Inference sovereignty caveat: The MCP server itself keeps API traffic EU-resident, but prompts and tool outputs are forwarded by the AI client to its LLM backend. Clients such as Claude Desktop and Cursor send inference to non-EU providers by default. Use AI Model Hub + a client supporting custom endpoints for fully sovereign workflows.
• Rate limiting: The IONOS CLOUD API enforces rate limits. If the agent makes too many rapid tool calls, the API returns HTTP 429. The binary surfaces this error to the AI client.
• No SLA for the binary: As a local open-source tool the MCP server carries no availability SLA. The wrapped IONOS CLOUD products (Compute Engine, Object Storage, etc.) carry their own SLAs.
• Build from source requires Go 1.21+: Pre-built binaries are available on the GitHub releases page. Building from source requires Go 1.21 or later.

Management Options

• GitHub releases page (pre-built binaries for all platforms, SHA-256 verified)
• Build from source: Go 1.21+ (go build)
• AI client config file (JSON mcpServers block with command and env)
• Claude Code CLI: claude mcp add ionoscloud /path/to/ionoscloud-mcp --env IONOS_TOKEN=...
• Project-scoped .mcp.json in a repo root for team-shared configuration