nfsn-cli

A modern command line interface and Python client for the NearlyFreeSpeech.NET API.

Covers the full documented API surface — Account, DNS, Email, Member and Site — plus a declarative, plan-and-apply workflow for DNS zones.

pip install nfsn-cli
nfsn init
nfsn dns list example.com

Why another one

The existing Python client, python-nfsn, was last committed in June 2018, targets Python 2.6–3.5, and depends on beanbag. It also predates several API members: balanceCash, balanceCredit, balanceHigh, replaceRR, and the sync property are all missing from it.

nfsn-cli is Python 3.11+, has no legacy baggage, ships a real CLI rather than a debug wrapper, and — importantly — encodes several API behaviours that are not in NFSN's documentation and that a naive client gets wrong. See MX records.

It reads ~/.nfsn-api if you already have one, so migrating costs nothing.

Configuration

nfsn init                      # writes ~/.config/nfsn/credentials, mode 0600
$EDITOR ~/.config/nfsn/credentials

Generate the API key in the member panel: Profile → Actions → Set/Change API Key.

NFSN_LOGIN=yourlogin
NFSN_API_KEY=...

Credentials are resolved in this order:

1. NFSN_LOGIN / NFSN_API_KEY environment variables
2. ~/.config/nfsn/credentials
3. ~/.nfsn-api (the JSON file NFSN's Perl library established)

NFSN rejects any request whose timestamp is more than 5 seconds from its own clock. nfsn-cli detects this from the response Date header and tells you to sync your clock, rather than reporting a misleading authentication failure.

Commands

nfsn dns      list, props, get, set, add, remove, replace, update-serial, sync,
              export, plan, apply
nfsn account  show, get, set-name, add-site, add-warning, remove-warning
nfsn email    forwards, set-forward, remove-forward
nfsn member   show
nfsn site     add-alias, remove-alias

Every mutating command prompts for confirmation; pass --yes to skip it. Read commands accept --json.

Declarative DNS

nfsn dns export example.com -o zone.yaml   # dump live state
$EDITOR zone.yaml
nfsn dns plan zone.yaml                    # diff, no changes made
nfsn dns apply zone.yaml --yes             # execute
nfsn dns apply zone.yaml --yes --wait      # ...and poll until fully propagated

apply without --yes is a dry run.

domain: example.com
records:
  - name: ""                                  # "" or "@" is the apex
    type: TXT
    data: v=spf1 include:amazonses.com ~all
    ttl: 3600
  - name: mail
    type: MX
    data: feedback-smtp.us-west-2.amazonses.com.
    aux: 10                                   # priority; see below

Safety model

By default apply only touches record sets the zone file actually names — a file listing three DKIM CNAMEs cannot delete your MX records, whatever else is in the zone. Anything outside those name+type pairs is reported as untouched and left alone.

--prune makes the file authoritative for the whole zone and deletes everything not listed. Run plan --prune first, every time.

Records NFSN owns (scope other than member) are never proposed for removal, even under --prune, and the client refuses to try.

TTL-only differences are not treated as changes; a delete/recreate cycle is not worth it.

One sharp edge: if the file declares an apex TXT for SPF and the apex also carries unrelated TXT records (domain verification tokens, say), those share the name+type pair and are considered managed — so they show up as removals. Read the plan.

MX records: read this

NFSN uses three different shapes for the same MX record across three verbs. None of this is documented; all of it was verified live against the API on 2026-08-17.

The consequences of getting this wrong are not subtle:

• Send the bare hostname to addRR and you create an MX record with no priority.
• Send the joined form to removeRR and the deletion silently 404s, so a "replace" leaves the old record in place and adds a second one beside it.

nfsn-cli handles the conversion for you: Record.wire_data produces the joined form for addRR, and remove_rr sends the split form. Zone files always use the split shape (data plus aux), matching what export gives you, and load_zone rejects an MX or SRV record with no aux rather than guessing a default.

replaceRR supports only A, AAAA and TXT, so it sidesteps the issue entirely — but that also means it cannot be used for MX or CNAME.

Python API

from nfsn_cli import Nfsn, NfsnTransport, Record
from nfsn_cli.config import load_credentials

credentials, _warnings = load_credentials()
with NfsnTransport(credentials) as transport:
    nfsn = Nfsn(transport)

    for record in nfsn.dns("example.com").list_rrs():
        print(record.display_name("example.com"), record.type, record.describe())

    print(nfsn.account("A1B2-C3D4E5F6").balance)
    print(nfsn.dns("example.com").sync)  # 0.0-1.0 propagation fraction

API coverage

Mirrors NFSN's API reference.

NFSN's Introduction page also names a Database object type, but its reference section documents no members, so it is not implemented.

Development

uv sync
scripts/install-hooks.sh
uv run pytest
uv run ruff check . && uv run ruff format --check .

See CONTRIBUTING.md — in particular the note on why mocked tests are not sufficient when changing how records are read or written.

• CHANGELOG.md — versions follow SemVer
• SECURITY.md — reporting vulnerabilities, and how your API key is handled
• CODE_OF_CONDUCT.md

License

MIT. Not affiliated with or endorsed by NFSN, Inc.