Model Context Protocol server bringing Hetzner to your AI agent
Hetzner MCP
Model Context Protocol server for full Hetzner Cloud + Storage API automation
Expose all official Hetzner operations as MCP tools so AI agents can manage servers, networking, load balancers, firewalls, volumes, DNS zones, and storage boxes from one server.
Install
pipx install hetzner-mcp
hetzner-mcp install
Then set your token and restart your MCP client:
export HETZNER_TOKEN="your_token_here"
Or configure it once via CLI (persisted local config):
hetzner-mcp auth set --token "your_token_here"
What It Does
hetzner-mcp loads official OpenAPI specs from Hetzner and exposes operations as MCP tools.
- Full Cloud API coverage:
https://api.hetzner.cloud/v1 - Full Storage API coverage:
https://api.hetzner.com/v1 - Dynamic tool generation from operation IDs
- Helper tools for discovery, search, and operation schema inspection
- Action polling helper for async action workflows
Core helper tools
| Tool | What it does |
|------|--------------|
| list_api_operations | List all operations with filters (domain/tag/method/query) |
| search_api_operations | Search operations by keyword |
| get_api_operation_details | Inspect full operation details and input schema |
| list_api_categories | List all API categories/tags with docs descriptions |
| get_api_category_details | Explain one category in depth with all endpoints inside |
| list_api_projects | Show configured project profiles and active credential context |
| set_active_api_project | Switch active project profile for agent execution context |
| wait_for_action | Poll cloud/storage actions until completion |
All API operations are also exposed directly as tools using the official operation ID names (for example create_server, get_action, create_storage_box).
For agent-friendly documentation, every endpoint and category also has dedicated guide tools:
- Endpoint guide tool pattern:
guide_<operation_id>- Example:
guide_create_server
- Example:
- Category guide tool pattern:
category_guide_<api_domain>_<category_slug>- Example:
category_guide_cloud_servers
- Example:
These guide tools include docs text from the OpenAPI documentation, explicit purpose, parameter explanations, and example tool arguments.
Docs-first execution lock (required)
This server enforces a docs-first workflow for endpoint execution:
- Call
guide_<operation_id>first for the endpoint you want to execute. - Then call the endpoint tool itself (for example
create_server).
If you skip step 1, execution is rejected with a docs_required error.
Unlocking is based on context freshness (interaction distance in the current session), not wall-clock time:
- Docs must be read before execution.
- Recently executed endpoints remain trusted while context is still fresh.
- After enough unrelated tool interactions (context drift), docs must be read again.
Full Coverage
Current generated operation coverage:
- Total operations: 221
- Cloud operations: 189
- Storage operations: 32
You can verify this locally:
python scripts/verify_operation_coverage.py
Example Prompts
"List all Hetzner operations related to firewalls"
"Create a CX22 server in fsn1-dc14 with my SSH key"
"Attach volume 12345 to server 67890"
"Create a load balancer and add target server 1001"
"Enable rescue mode on server 123 and wait for action completion"
"Create a storage box and reset its password"
"Show operation details for update_storage_box_access_settings"
Authentication
You can configure auth in two ways:
- Environment variables (recommended for CI/ephemeral environments)
- Local CLI config file (recommended for local workstation use)
Environment variables (highest precedence):
HETZNER_TOKENfor both Cloud and Storage APIsHETZNER_CLOUD_TOKENto override cloud auth tokenHETZNER_STORAGE_TOKENto override storage auth tokenHETZNER_PROJECTto choose one configured local project profile for this session
Local CLI config examples:
# set default token
hetzner-mcp auth set --token "your_token_here"
# set per-domain overrides
hetzner-mcp auth set --cloud-token "cloud_token" --storage-token "storage_token"
# inspect effective token sources (env/file/unset)
hetzner-mcp auth show
# open full local config in your editor
hetzner-mcp config edit
Multi-project profile examples:
# create per-environment profiles
hetzner-mcp project add prod --description "Production Hetzner" --token "prod_token" --activate
hetzner-mcp project add staging --description "Staging Hetzner" --token "staging_token"
# see which profile is active and what each one is for
hetzner-mcp project list
# switch active profile
hetzner-mcp project use staging
Config file location:
~/.config/hetzner-mcp/config.json(macOS/Linux)%APPDATA%\\hetzner-mcp\\config.json(Windows)- Override path with
HETZNER_MCP_CONFIG_PATH
Optional runtime controls:
HETZNER_CLOUD_BASE_URLHETZNER_STORAGE_BASE_URLHETZNER_TIMEOUT_SECONDSHETZNER_MAX_RETRIESHETZNER_BACKOFF_BASE_SECONDS
Supported MCP Clients
Auto-configuration is included for:
| Client | Auto-config | |--------|-------------| | Claude Desktop | Yes | | Claude Code | Yes | | Cursor | Yes | | Cline | Yes | | Windsurf | Yes | | OpenCode | Yes |
Run:
hetzner-mcp install
Manual Configuration
{
"mcpServers": {
"hetzner-mcp": {
"command": "hetzner-mcp-server",
"args": []
}
}
}
CLI Commands
| Command | Description |
|---------|-------------|
| hetzner-mcp status | Show effective runtime config + registry + client status |
| hetzner-mcp doctor [--json] | Print detailed diagnostics |
| hetzner-mcp server run [--refresh-specs] | Run stdio MCP server |
| hetzner-mcp update | Refresh specs and re-apply client integration |
| hetzner-mcp client install | Configure supported MCP clients |
| hetzner-mcp client status | Show client config installation state |
| hetzner-mcp client repair | Re-apply configuration entries |
| hetzner-mcp client uninstall | Remove MCP config entries |
| hetzner-mcp auth set ... | Configure API keys directly from CLI |
| hetzner-mcp auth show | Show token status and source |
| hetzner-mcp auth clear [--all] | Clear stored token entries |
| hetzner-mcp project add/list/show/use/remove | Manage multiple project credential profiles |
| hetzner-mcp config show | Show stored + effective config |
| hetzner-mcp config get/set/unset <key> | Read/write persisted config keys |
| hetzner-mcp config edit | Edit persisted config file in $EDITOR |
Legacy aliases still work: install, repair, uninstall, diagnose.
Development
git clone https://github.com/devinoldenburg/hetzner-mcp.git
cd hetzner-mcp
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
# Validate quality
ruff check .
mypy src
pytest
# Refresh specs and inspect counts
python scripts/sync_specs.py
python scripts/verify_operation_coverage.py
Security Notes
- Never commit API tokens.
- Tokens are read from environment variables.
- Server logs are routed to stderr to keep stdio JSON-RPC clean.
- Network retries are limited and capped.
Troubleshooting
ModuleNotFoundError: hetzner_mcp in local scripts
- Install editable package:
pip install -e .
No operations listed in MCP client
- Run
hetzner-mcp status - Verify config file includes
hetzner-mcp - Restart the client process after installation
Auth failures (401 unauthorized)
- Ensure
HETZNER_TOKENor domain-specific token is exported in the client runtime environment
Rate limiting (429)
- The client retries transient failures automatically; reduce request burst and retry later
License
MIT