SSH commands

The agent sends a command. Sallyport signs in with your key or password, runs it and returns the output. The agent never sees either.

Add the host

• Open Connect an SSH host. From Credentials, it is Add a credential, then SSH host under Credential source.
• Enter a Connection name. Agents see it, and the host’s tool is named after it.
• Enter the Public hostname or IP address, the SSH port and the SSH user. The port is 22 unless you change it.
• Pick the Workspace if you have more than one, and the Environment: Live or Test.
• Under Sign in with, keep Private key and paste the OpenSSH private key. If the key has a passphrase, enter it in Key passphrase (optional).
• A server that takes no key, such as shared hosting, a router or an appliance: choose Password and enter the SSH password. Sallyport answers both the password and the keyboard-interactive prompt with it.
• Press Save SSH connection. The key is write-only, and nobody can read it back.

Sallyport checks the address when you save. A private address, or one that points at Sallyport itself, is refused.

Check the server’s identity

Host-key verification (optional) offers two choices.

The connection page shows the fingerprint under Host key verification. If the server’s key changes, calls stop with host_key and no command runs.

Confirm the new key with the server’s owner first. Then open the connection’s page, press Replace pinned key under Host key verification, paste the new SHA256: fingerprint and press Pin this key. Approvals given for the old key stop counting. A script does the same with a console key (REST API). GET /v1/ssh-hosts lists your hosts with their id. PATCH /v1/ssh-hosts/{id} with known_host_key pins the new key.

export SP_CONSOLE_KEY='sp_ck_…'

curl --fail-with-body -H "Authorization: Bearer $SP_CONSOLE_KEY" \
  https://mcp.sallyport.cloud/v1/ssh-hosts

curl --fail-with-body -X PATCH -H "Authorization: Bearer $SP_CONSOLE_KEY" \
  -H 'Content-Type: application/json' \
  -d '{"known_host_key": "SHA256:<new fingerprint>"}' \
  "https://mcp.sallyport.cloud/v1/ssh-hosts/<host-id>"

Choose which commands wait for you

The new connection goes to every active agent of its workspace and environment. Commands run without asking by default.

To approve commands, open the agent in Agents, find the connection in its Access list and turn on Ask every call. Or set one class at a time:

• Read: plain read commands from a fixed list, such as ls, cat, tail, git status, docker ps and kubectl get.
• Destructive: commands such as rm, dd, mkfs, kubectl delete and docker system prune.
• Write: everything else. A pipe, ;, a redirect, $, a wildcard, sudo or sh -c makes a command a write, even when it only reads.

The approval card shows the host and the exact command. That command, byte for byte, is what runs.

Run a command

The agent keeps its Sallyport MCP connection. list_access names the host’s tool, which ends in _exec. The tool takes these arguments:

{
  "cmd": "df -h",
  "intent": "Check free disk space before the deploy"
}

Once the agent has an SSH host, it also gets ssh_exec. It takes the same arguments plus host. A host outside the agent’s list gives a link where you grant access.

The same host gives the agent its files too. _sftp_read and _sftp_write list, read, write, rename and delete them over SFTP with this key. See Files over SFTP.

Read the result

• The answer carries exit_code, stdout and stderr. Output that is not UTF-8 arrives as stdout_base64 or stderr_base64.
• Long output is cut at max_inline_bytes, and fetch_body reads the rest. One command may print up to 512 MiB.
• A command gets no terminal and no input. Use flags that skip prompts, such as apt-get -y.
• Open the call in Activity to see SSH command result: the target, Exit code, Host-key fingerprint and output sizes.