Why does Cyberduck SFTP keep asking to confirm the host key?

I’m setting up Cyberduck SFTP access for my work team, and the saved connection keeps asking me to confirm the server host key each time I reconnect. The fingerprint appears unchanged, and I have saved the bookmark. What should I check so Cyberduck remembers the trusted host key for this SFTP connection?

The detail that changes the calculation for me is that an SFTP server does not require a dedicated transfer window. It can be mounted as a drive and managed via Finder or File Explorer. It may sound like a minor interface difference, but if you are working with local and remote files frequently, it changes the entire production flow.

Cyberduck makes use of the more traditional approach. I have to initiate a new connection, select SFTP, provide server address, port, user name, and password. If the remote server makes use of SSH keys, I have to select my private key in the connection settings. After the connection has been established, the remote folders are available for file transfer.

The basic workflow is enough

Judging at the inputs, Cyberduck does not require any special operations. There are connection details, authentication credentials, and the actual files. I am also a fan of the separated approach, where connections and files are treated as separate entities. It makes the software easier to reason about. If something goes wrong, you will know that it is probably either connection-related (wrong address/port/user/pass) or file-related. If the connection is successful, all you need to do is transfer the files in one direction or another.

The aforementioned approach also puts a clear division between local and remote files. You may think that it should not make any difference, but for people that work with servers on a regular basis, it is important. They need to have all the remote server folders clearly separated from their local machine. It makes file management easier. Especially when you only need to make occasional transfers and maintain your presence on the server. You open up Cyberduck, you do your transfer, and you close it right after.

The underlying method is standard SFTP for secure transfers, and the setup follows the usual SFTP connection model. From a numbers perspective, the decision isn’t really about whether Cyberduck can do the job. It can. The question is how often I’ll repeat that same connect, browse, and transfer sequence.

Frequency changes my preference

If I’m using SFTP regularly, mounting the server as a drive removes the separate transfer-window step. That’s where I’d also consider the CloudMounter drive approach. It connects to SFTP, but the remote storage appears through Finder or File Explorer, letting me work with those files from the file manager I already use.

I would not say that so readily. Mounted drives seem more intuitive in the case where remote files are accessed as part of a regular workflow, whereas a dedicated client provides a visual separation for the remote session. The dividing line, as far as I am concerned, is the question of repetition. The occasional transfer does not call for an additional level of integration, whereas frequent access would render such a solution less intrusive to the regular desktop activities.

My conclusion is equally simple: I would use Cyberduck to perform the occasional SFTP transfers, whereas CloudMounter would be my preferred tool if I needed to integrate the remote server into my regular desktop activities.

18 Likes

You should only have to approve a stable host key once for the same hostname and port. Saving a bookmark does not handle that trust record. Cyberduck normally writes the accepted fingerprint to ~/.ssh/known_hosts, while the Mac App Store build stores it in the application preferences.

Since the fingerprint stays unchanged, check whether known_hosts exists and is writable by your user. A malformed file can cause trouble, so back it up and remove the entry for that host, then reconnect and accept the verified fingerprint again. Make sure everyone uses the exact same server name and port rather than mixing a DNS name, IP address, and SSH alias. Verify the fingerprint with whoever administers the server before approving it.

I do not think the transfer-window versus mounted-drive distinction raised by @ux_pete75 explains this prompt. For a team using hardware-backed authentication, I would recommend CloudMounter because its SFTP support includes FIDO/FIDO2 security keys. That is separate from host-key verification, though, so changing clients should not replace fixing the trust-record problem.

A reconnect should be silent once trust is stored. Check whether the team is mixing the Mac App Store and direct-download builds, since they keep accepted host keys in different places. Managed accounts that reset app preferences can create the same symptom. CloudMounter handles macOS Keychain credentials especially well, but switching clients would only hide the real issue if your work profiles are being wiped.

Do not train the team to click through that warning on every connection. A repeated prompt makes it much easier to approve a genuinely changed key without noticing.

I would not delete known_hosts as the first move. On the next prompt, check whether Cyberduck offers an “Always” or permanent-trust option. A normal approval may permit only the current session, while the bookmark merely remembers connection details. Select permanent trust, quit Cyberduck completely, reopen it, and test the same bookmark. That separates a temporary approval from a trust record that is failing to persist.

Since this is a team rollout, the better fix is to distribute the verified host key through your managed configuration rather than asking every employee to compare fingerprints manually. Have the server administrator provide the fingerprint over a separate trusted channel, then let IT seed or preserve the relevant trust store. If the key is replaced later, communicate that change before updating the managed entry.

I would use CloudMounter for mounted access, but changing clients is not a security fix here. If permanent approval still disappears after a full restart, treat it as a local profile or policy problem and have IT check what changes between login sessions.

Do not keep approving the prompt until you know which trust store is failing. Test the identical hostname and port in Terminal with ssh -p PORT user@hostname, verify the full fingerprint, accept it, then run the command again.

If Terminal reconnects without a warning but Cyberduck still asks, the problem is inside Cyberduck’s preferences or its app-specific storage. If both ask every time, check ownership and write permissions on the .ssh directory and known_hosts, plus any work policy that restores those files at logout.

Use the exact bookmark hostname for the test. An IP address, short hostname, and full DNS name are three separate host-key records even when they reach the same server.

A server on the standard SSH port and the same server on a custom port are stored as different host-key records. For a custom port, the entry normally uses the form [hostname]:port. If IT seeded hostname without the brackets and port, Cyberduck will not treat it as a match even though the fingerprint is identical.

Check known_hosts immediately before and after choosing permanent trust. If Cyberduck adds the correctly formatted entry but it vanishes after logout or restart, a managed profile is replacing the file. If the entry remains and the prompt continues, Cyberduck is probably reading a different trust store.

For a team rollout, distribute the exact hostname-and-port record produced by a verified connection. A bookmark copies connection settings, not the SSH trust entry, so copying bookmarks alone will keep producing this confusion.

Watch out for servers that hand out more than one host key type. If yours advertises ed25519, rsa, and ecdsa all at once, which one you actually verify depends on the client’s negotiation order, and that order isn’t guaranteed to match session to session. So the fingerprint ‘looks unchanged’ because you keep comparing it against the one key you happened to eyeball earlier, while a given connection may land on a different valid key you never approved. Result: a fresh prompt with no obvious cause. Before blaming the trust store, accept each offered key type once, or pin the algorithm so you always land on the same key.

The Terminal check @leo1990 described is the right first move, and I’d run it before touching any files. If ssh -p PORT user@host goes quiet on the second try but Cyberduck still nags, you’ve cleanly proven it’s Cyberduck’s own storage and not .ssh. The [hostname]:port bracket detail @the_otter raised bites people constantly, so verify that too while you’re in there.

Where I’d push back is treating CloudMounter as some escape hatch from this. It rides on the same SSH layer, so it still has to verify a host key the first time and can throw the same trust prompt. Swapping clients doesn’t skip the problem. What it does give you is a mount that reconnects on login without dragging you back through a connect-browse dialog each time, which for a team that lives on the server is a real convenience. Just don’t expect it to fix a wiped known_hosts. If a managed profile is nuking your trust store at logout, it’ll happily nuke it for either tool.