hetzner-dns-mcp

A Model Context Protocol (MCP) server for managing DNS zones and records via the Hetzner Cloud API.

Lets MCP clients like Claude Code, Claude Desktop or Codex manage your Hetzner DNS: list, create, update and delete zones and RRSets (record sets), import/export zone files, manage protection and TTLs, and track asynchronous zone actions.

Twenty-two tools is the ceiling, not the floor: HETZNER_ALLOW_TOOLS=essential registers a curated eight instead, and a model picks the right tool far more reliably from eight than from twenty-two — see choosing which tools load.

Note: This server targets the current DNS API that is part of the Hetzner Cloud API (api.hetzner.cloud). The legacy DNS API (dns.hetzner.com) was shut down in May 2026 and is not supported.

What makes it different

Zones, record sets and the actions behind them. Listing, creating, updating and deleting zones and RRSets, importing and exporting BIND zone files, changing TTLs and protection — and following the asynchronous zone actions Hetzner queues behind a change, rather than reporting success and leaving you to guess.

Nothing upstream is trusted. Record values, comments and zone files come back wrapped as untrusted data, secret-looking keys are redacted, oversized values are truncated and HTML error pages are dropped instead of pasted into the context.

Destructive calls ask a person. Deleting a zone or replacing a record set raises a real dialog through MCP elicitation. Where the client cannot show one, the call is refused and carries a random single-use token that only ever appeared in a previous tool result — so nothing hidden inside a DNS record can mint it.

Requirements

• Node.js ≥ 22
• A Hetzner Cloud API token for the project that holds your DNS zones — create one in the Hetzner Cloud Console under your project → Security → API tokens. Use a read & write token for full functionality (a read-only token limits you to the read tools).

Configuration

Configuration is provided via environment variables:

Without a token the server still starts and lists its tools (so registries and inspectors can introspect it), but every tool call fails with setup instructions instead of reaching the API.

HETZNER_API_TOKEN and HETZNER_API_BASE_URL are deleted from the process environment once they have been read, so a later crash report or diagnostic dump cannot expose the token.

Choosing which tools load

HETZNER_ALLOW_TOOLS and HETZNER_DENY_TOOLS take comma-separated tool names; a trailing * matches a whole family. essential is a curated preset — list_zones, get_zone, list_rrsets, get_rrset, create_rrset, set_records, delete_rrset and export_zonefile — which covers reading a zone and changing a record without the rare, catastrophic tools.

HETZNER_ALLOW_TOOLS=essential
HETZNER_ALLOW_TOOLS=list_*,get_zone,set_records
HETZNER_DENY_TOOLS=delete_zone,import_zonefile

An entry that matches no tool aborts startup and names it, so a typo cannot silently hide a tool — an absent tool is not something anyone traces back to an environment variable. A filtered tool is never registered, so it is absent from tools/list and unknown to tools/call alike, exactly like a write tool under HETZNER_READ_ONLY.

If you run several of these servers at once, mcp-hub is the other answer — its /hub endpoint replaces every server's tools with six meta-tools.

Installation

Claude Code

claude mcp add hetzner-dns -s user \
  -e HETZNER_API_TOKEN=your-token \
  -- npx -y hetzner-dns-mcp

Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "hetzner-dns": {
      "command": "npx",
      "args": ["-y", "hetzner-dns-mcp"],
      "env": {
        "HETZNER_API_TOKEN": "your-token"
      }
    }
  }
}

Codex

Add to your ~/.codex/config.toml:

[mcp_servers.hetzner-dns]
command = "npx"
args = ["-y", "hetzner-dns-mcp"]
env = { HETZNER_API_TOKEN = "your-token" }

From source

git clone https://github.com/ni-c/hetzner-dns-mcp.git
cd hetzner-dns-mcp
npm install
npm run build
# then use `node /path/to/hetzner-dns-mcp/dist/index.js` as the command

Docker

Multi-arch images (amd64/arm64) are published to GHCR with an SBOM and build provenance:

docker run -i --rm -e HETZNER_API_TOKEN=your-token ghcr.io/ni-c/hetzner-dns-mcp

The image talks MCP over stdio, so clients need docker run -i (no port is exposed):

{
  "mcpServers": {
    "hetzner-dns": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "HETZNER_API_TOKEN",
        "ghcr.io/ni-c/hetzner-dns-mcp"
      ],
      "env": {
        "HETZNER_API_TOKEN": "your-token"
      }
    }
  }
}

Through mcp-hub

A client that cannot spawn a local process — ChatGPT connectors, Claude on the web, Cursor, LibreChat — reaches hetzner-dns-mcp through mcp-hub: one container serves many stdio MCP servers over Streamable HTTP, with an OAuth 2.1 login behind a single password and long-lived tokens for the clients that cannot do OAuth. Its /hub endpoint puts every server behind six meta-tools, so one connector reaches all of them without N×tool schemas in the model's context, and it speaks both protocol revisions — a question this server asks travels through it to the person at the far end.

Its /config/mcp.json uses Claude Code's format, so the entry is the one you already have:

{
  "mcpServers": {
    "hetzner-dns": {
      "command": "npx",
      "args": ["-y", "hetzner-dns-mcp"],
      "env": {
        "HETZNER_API_TOKEN": "your-token",
        "HETZNER_ALLOW_TOOLS": "essential"
      },
      "denyTools": ["delete_*"]
    }
  }
}

allowTools and denyTools there are the hub's own per-server filter, which is not the same thing as *_ALLOW_TOOLS in env — the difference, and the mistake it invites, are in the client guide.

Tools

Every tool declares an outputSchema and answers with structuredContent alongside the text block, so a client can use the result without parsing prose. The <untrusted-data> fence stays in the text — it is the readable presentation of the same marker — and the structured half carries untrusted: true and source: "hetzner-cloud-api" as fields. Every tool carries them: record values, comments, labels and zone files are written by whoever controls the zone, and no tool here answers with anything else.

The API documents are described as open objects with the top-level keys the spec guarantees. The API is not this server's to promise, and the SDK validates each result against its schema before it goes out — a strict shape would turn a field Hetzner adds into a tool that fails outright.

An over-budget result is now an error. It used to be cut off mid-document and say so, which is fine for a text block and impossible for structuredContent: the two channels have to carry the same value, and a document sliced mid-string does not parse.

Zones

RRSets (record sets)

👤 asks a person through MCP elicitation · falls back to a two-call confirm_token where the client cannot show a dialog.

Actions

Safety

A person is asked, not just told. Every irreversible tool — delete_zone, delete_rrset, import_zonefile, set_records, remove_records, change_primary_nameservers, and change_zone_protection/change_rrset_protection when they remove protection — raises a real dialog through MCP elicitation where the client supports it. The model cannot answer it on its behalf, and nothing happens until an answer comes back.

Three more ask conditionally, because in DNS the dangerous act is making a claim rather than withdrawing one: create_rrset and add_records for a name or type that decides who answers (NS, DS, MX, CNAME, CAA, TLSA, SVCB, HTTPS, SRV, the apex @, a wildcard), and create_zone when the call carries primary_nameservers or a zonefile — the same payloads change_primary_nameservers and import_zonefile are guarded for. The confirmation binds every field the call will write, TTL and labels included, so a token issued for one TTL cannot redeem another.

Where the client cannot show a dialog, the tool refuses its first call and returns a random, single-use token valid for five minutes; the second call must repeat the identical arguments and pass it:

1. set_records(zone: "example.com", name: "www", type: "A", records: [{value: "198.51.100.1"}])
   → error: Refusing to replace the records … confirm_token: "3f9c…"
2. set_records(…same arguments…, confirm_token: "3f9c…")
   → executed

Deliberately not a boolean the model can set on its own: the token exists only in a previous tool result, so an instruction hidden in a TXT record or a zone-file comment cannot manufacture one. But it proves the call was made twice with the same arguments and nothing more, and the fallback text says so rather than implying somebody approved.

Either way the approval is bound to what it is about. For set_records, remove_records and import_zonefile that is a hash of the exact payload: an approval for ["198.51.100.1"] will not write ["198.51.100.66"].

ELICITATION=false takes the fallback path deliberately, for a scheduled job or a test harness. It never removes the guard. See Asking a person.

The dialog is a guard rail, not a security boundary — the boundary is the permission prompt of your MCP host. Do not auto-approve these tools.

Untrusted upstream data. Everything the API returns is wrapped in an <untrusted-data> envelope, because record values, comments, labels and zone files are written by whoever controls the zone. Confirmation messages never quote that content; they report counts and TTLs only. Keys that look like secrets (tsig_key, token, secret, password) are redacted from results, oversized values are truncated, and HTML error pages from an intermediate proxy are dropped rather than pasted into the model's context.

Least privilege. Set HETZNER_READ_ONLY=true to register only the seven read tools — the write tools then do not exist on the protocol at all, rather than failing at call time. Combine it with a read-only Hetzner token for a genuinely read-only setup.

Other guarantees.

• Tools carry MCP annotations (readOnlyHint, destructiveHint, idempotentHint) so hosts can apply appropriate permission policies.
• HETZNER_API_BASE_URL must be an https URL (http is only accepted for localhost) and must not contain credentials; a warning is printed when a non-default host is configured, because the API token is sent there.
• Zone identifiers and RRSet names are validated against a strict character set — no slashes, no percent signs, no bare . or .. — so a request cannot escape the intended API path; requests never follow redirects, so the Bearer header cannot be replayed elsewhere.
• TSIG keys passed to create_zone/change_primary_nameservers become part of the conversation context and client transcripts — treat them as secrets and rotate them if in doubt.
• Hetzner-side resource protection is honored: protected zones/RRSets return an error with a hint to the corresponding change_*_protection tool.

Not exposed, on purpose

Only the Hetzner Cloud DNS API. The standalone dns.hetzner.com API was shut down in May 2026 and is not implemented — scripts still using it are broken for reasons that have nothing to do with this server.

HETZNER_API_BASE_URL is not a provider switch. It exists to point at a local mock during development, and is validated (HTTPS only, no credentials in the URL, a warning on a non-default host) precisely because your token is sent there. The request shapes are Hetzner's.

Safety

• DNS records are edited by whoever holds the token, so upstream values, comments and zone files are marked as untrusted data — to be reported, never followed, in both channels and behind a fence the data cannot close. Keys are redacted by the suffix of their normalised name (password, secret, token, apikey, privatekey, passphrase, tsigkey), values over 4 000 characters are truncated, control characters are stripped, and an HTML error page is dropped rather than pasted into the context.
• Deleting a zone or replacing a record set asks a person: a real dialog through MCP elicitation, bound to the exact target. Where the client cannot show one, the call is refused and carries a random single-use token that only ever appeared in a previous tool result.
• HETZNER_READ_ONLY=true registers the seven read tools and nothing else — a write tool is then absent from tools/list, not refused when called.
• Requests are hardened rather than trusted: zone identifiers are pattern-matched so a path cannot escape, redirects are refused so the Authorization header never follows one, every request times out after 30 seconds, request bodies are assembled from named fields only, and the base URL is validated (HTTPS, no credentials) before the token is ever sent to it. The token's own shape is checked before it reaches the HTTP layer, which would otherwise quote it back in an error; response bodies are read under a ceiling with the status decided first; and every response is shape-checked rather than cast, so a proxy's answer produces a readable result rather than a schema violation with no cause.

Documentation

The full guide, tool reference and security notes live at hetzner-dns-mcp.ni-c.de (source in docs/).

Development

npm install
npm run build     # compile TypeScript to dist/
npm test          # run the vitest test suite
npm run lint      # oxlint + prettier check

Releasing

1. Bump the version in package.json (npm version X.Y.Z --no-git-tag-version).
2. Rename the [Unreleased] section in CHANGELOG.md to the new version.
3. Commit, then tag and push: git tag -a vX.Y.Z -m "vX.Y.Z" && git push origin main vX.Y.Z.

The release workflow then runs the tests, publishes to npm via Trusted Publishing (with provenance), creates a GitHub release with the notes from the CHANGELOG and updates the entry in the official MCP Registry (io.github.ni-c/hetzner-dns-mcp, via GitHub OIDC).

Releasing

Releases are tag-driven. Bump package.json, move the [Unreleased] notes in CHANGELOG.md under the new version, commit, then:

git tag -s vX.Y.Z -m "vX.Y.Z"
git push origin main vX.Y.Z

The release workflow publishes to npm via Trusted Publishing (OIDC, with provenance), pushes the multi-arch container image to GHCR, creates the GitHub release from the CHANGELOG section, and updates the entry in the official MCP registry.

Contributing

Issues, discussions and pull requests are welcome — see CONTRIBUTING.md. For vulnerabilities please use private reporting rather than a public issue; the policy is in SECURITY.md.

License

MIT © Willi Thiel