Codex + xCloud

Secure sites and servers with Codex on xCloud

Securing sites and servers with Codex means telling OpenAI's coding agent, from your terminal, to read vulnerability findings and server protections on xCloud, and to change a firewall rule or a ban only after you approve it.

Skill: xcloud:servers
Toolsets: vulnerabilities, servers, sites, wordpress-actions
Free with every xCloud account

Set up Codex All Codex guides

YouShow Fail2Ban bans on Frankfurt and unban 203.0.113.10.

servers_fail2ban_banned-ipsread-only

AgentFourteen addresses are banned. 203.0.113.10 is on the list. I will unban it on the Frankfurt server, which lets it reach SSH and the web again. Go ahead?

YouYes.

servers_fail2ban_unbanneeds approval

AgentUnbanned. Thirteen addresses remain on the ban list.

Last updated: October 2026. Keep a human in the loop: xCloud stops and asks before anything that creates, deploys, updates, reboots, deletes or buys.

Setup

How Do You Set Up Codex to Secure sites and servers on xCloud?

Connect Codex once; every job on this account uses the same connection. Then ask in plain words.

1
Add the xCloud MCP server
Run this in your terminal. This form comes from xCloud's connect guide; OpenAI's Codex page documents codex mcp add for local commands, so if your version rejects --url, use the config.toml step below.
Terminal
```
codex mcp add xcloud --url https://app.xcloud.host/mcp
```
2
Or write it in config.toml
Codex reads ~/.codex/config.toml. Add a table for the server with its url. A Streamable HTTP server needs only the url, and the same file is where you set a bearer_token_env_var if you use an API key instead of OAuth.
TOML
```
[mcp_servers.xcloud]
url = "https://app.xcloud.host/mcp"
```
3
Sign in and check
Run the login command, approve the teams and choose Read-only or Full access in the browser, then type /mcp in Codex to see xcloud and its tools. Optional: Codex is also listed for the xCloud Agent Plugins package, which adds the xCloud skills.
Terminal
```
codex mcp login xcloud
```

Check it worked

Then ask Codex for the job itself, for example:

Prompt

Rank the sites on this team by open vulnerabilities and name the top three findings.

In practice

How Does Security Work from Codex?

Codex lives in a terminal session next to your repository, which makes it a handy place to run a quick security check between tasks. After codex mcp login xcloud, the /mcp command lists the xcloud server and its tools, and you can open with a plain request. Ask which sites on the team have open vulnerabilities and Codex calls the team-wide rollup, then prints a compact ranking. Ask about one site and it reads the counts by severity and the findings, with the plugin or theme named. You do not copy anything out of a dashboard, and the answer lands in the same scrollback as the rest of your work.

Codex is direct about what it will and will not do on its own. Reads run straight away. A rescan is a routine action that runs without a stop, but because it finishes asynchronously Codex checks the count again before it reports a result. Updating a plugin or theme to clear a finding always stops for your approval, and Codex can offer a backup first. When you ask about a server it reads the firewall rules, the Fail2Ban bans, the SSH restriction status and the sudo users and summarises them. It can also write the summary into a file in the repository, such as an incident note, if you ask, which is something a chat window cannot do.

Changes to the perimeter follow a fixed pattern. Before Codex creates, disables or deletes a rule, bans an address, unbans one or whitelists xCloud's own IPs, it names the exact server and the effect and waits for your yes, and it explains the lockout risk if the change touches SSH. It will not delete the default SSH rule on its own. Temporary sudo users are for one job and removed afterwards. For dashboard-only protections, such as the AI bot blocker and the 7G and 8G firewall, Codex says so and gives you the path and the dashboard link rather than saying it switched them on.

Codex specific: A recurring check in a script, such as codex exec in a scheduled job, has no one to approve a change, so give it a read-only token. Set bearer_token_env_var in config.toml to the name of a variable that holds a token with the mcp:invoke scope plus read:servers and read:sites and no write abilities, keep the token in your shell or secret store and never write it into config.toml itself, then reserve interactive sessions with Full access for the days you intend to change a rule.

What xCloud does for security

xCloud reads vulnerability findings for one site or across your whole team, starts rescans and ignores or restores a finding. On servers it manages firewall rules, Fail2Ban bans, the SSH restriction state and sudo users. Protections such as the AI bot blocker, the 7G and 8G firewall and per-site IP rules stay dashboard switches.

Start with the team-wide rollup. For a fleet question the agent reads the vulnerability rollup across every site on the team in one call and sorts it worst first. For one site it reads the findings and the counts by severity instead.
Rescan and triage findings. A rescan runs without a confirmation stop and finishes asynchronously, so the agent checks the count again afterwards. Findings carry a severity and a source, and severity is the triage key. A false positive can be ignored with a reason and restored later.
Fix what the finding points to. Most findings are fixed by updating a plugin, a theme or WordPress core. The agent proposes the updates, takes a backup first if you want one, and waits for your approval, because updating plugins or themes always stops for confirmation.
Review the server's perimeter. The agent reads firewall rules, the Fail2Ban banned IP list, the SSH restriction status and the sudo users. It can add or disable a rule, ban or unban an address and whitelist xCloud's own IPs after you confirm the server and the effect.
Report what is dashboard-only. AI bot blocking, the 7G and 8G firewall and a single site's IP allow and block list are dashboard switches. The agent reads what it can, such as the 7G and 8G logs, and gives you the dashboard path and the site's dashboard link for the rest.

Reference

Security Settings and Limits on xCloud

The facts Codex works within when it secures sites and servers. Where a row names the dashboard, that step stays yours to take there.

Setting or limit	What applies
Vulnerability findings	Per-site list and counts by severity (critical, high, medium, low), plus a team-wide rollup across every site you can reach
Rescans	Triggering a rescan is a routine action that runs without a confirmation stop. It is asynchronous, so the agent polls the count or the site events
Ignore and restore	A finding can be ignored with a reason and un-ignored later. Both are changes to the finding, not to the site
Firewall rules	List, create, enable, disable and delete server rules by name, protocol (tcp or udp), traffic (allow or deny), port and optional source IP
Fail2Ban	List banned IPs, ban one or more addresses and unban an address. Adding addresses to Fail2Ban's own ignore list is an SSH step described in the Fail2Ban guide
Whitelisting	xCloud's infrastructure IPs and the caller's IP can be whitelisted. The SSH restriction status says whether SSH is limited to trusted addresses
Sudo users	List, create or update, and delete OS-level privileged accounts. Private keys are never returned, and a temporary account is available for short jobs
7G and 8G firewall	Switched on in the dashboard. Their logs are readable through the nginx log type of the site's access log read
AI bot blocker	A dashboard toggle: Site, Tools, Nginx and Security on Nginx, Tools, Security on OpenLiteSpeed
Per-site IP rules	Dashboard-only: Site, Tools, IP Management. The API lists a site's rules but does not change them; server-wide blocks are on the API
Automated security updates	xCloud installs security updates on servers it provisions, and Automatic Reboot restarts the server only when a reboot is required. Both are configured in the dashboard
Security headers	The API lists a site's custom Nginx configuration. Adding headers is a dashboard or SSH step, described in the security headers guide

Rules Codex has to follow

Firewall, Fail2Ban and sudo-user changes name the exact server and the effect before they run, and they wait for your yes. Locking out SSH is the risk the agent explains first.
The agent never disables SSH or a database service to tidy up a ruleset, and it does not delete the default SSH rule on its own.
Updating a plugin or theme to fix a finding stops for your approval, and the agent offers a backup first.
Ignoring a finding needs a reason you give. The agent records it and tells you how to restore it.
AI bot blocking, the 7G and 8G firewall, per-site IP rules and automated update settings are dashboard steps. The agent says so, gives the path and the dashboard link, and never claims it changed them.
A temporary sudo user is for a specific job and is removed when the job ends.

Example prompts

What Can You Ask Codex to Do for Security?

Type these as written and swap in your own repository, site and server names. Reads and routine actions such as backups, cache purges, PageSpeed scans and vulnerability scans run straight away; creating, deploying, updating, rebooting, deleting, buying or starting a broken-link scan stops and asks first.

Prompt

Rank the sites on this team by open vulnerabilities and name the top three findings.

Prompt

Rescan blog.example.com, wait for it to finish and list what remains open.

Prompt

Show Fail2Ban bans on Frankfurt, unban 203.0.113.10 and write a short incident note to notes/ban-review.md.

Prompt

Show the vulnerability count for every site on this team, sorted worst first, and name the top three findings.

Prompt

Run a vulnerability scan on the shop site and tell me what it finds.

Prompt

List the findings on the blog site and ignore the false positive in the contact form plugin, with the reason.

Prompt

Update every plugin on the shop site, then run a vulnerability scan, and stop if anything is still open.

Prompt

List the firewall rules on the Frankfurt server and tell me whether SSH is restricted to trusted IPs.

Prompt

Show the IPs Fail2Ban has banned on the Frankfurt server, and unban 203.0.113.10.

Prompt

Whitelist the xCloud IPs on the Frankfurt server.

Prompt

Which sudo users exist on the Frankfurt server? Flag any that look temporary or unused.

Codex and Security: Frequently Asked Questions

What people ask before they let Codex secure sites and servers through xCloud.

Can Codex run a security check from a script?

Yes, with a token supplied through bearer_token_env_var and an agreed Read-only scope, a non-interactive Codex run can read findings, rules and bans and report them. Changes need a person to approve them, so keep unattended runs read-only.

Does Codex see my sudo users' private keys?

No. xCloud lists sudo users, creates or updates them and deletes them, but it never returns private keys. Codex can flag accounts that look temporary or unused and remove one only after you confirm the server and the account.

Can an agent show vulnerabilities across all my sites at once?

Yes. xCloud has a team-wide rollup that lists findings for every site you can reach, which the agent sorts worst first. For one site it reads the findings and the counts by severity, and it can start a rescan.

Can the agent ban an IP address on my server?

Yes. It can ban addresses in Fail2Ban, list the banned IPs and unban one, and it can create, disable and delete server firewall rules. It confirms the server and the effect first, so you do not lock yourself out.

Can the agent turn on the 8G firewall or the AI bot blocker?

No. Both are dashboard switches. The agent can read the 7G and 8G firewall logs through the site's nginx logs, and it tells you where the switches are: the site's Tools section. It does not claim to have changed them.

Can the agent block an IP for just one site?

Not from here. Per-site IP allow and block lists are managed in the dashboard under Site, Tools, IP Management, and the agent can only list what is there. Server-wide blocks, through firewall rules and Fail2Ban, are on the API.

Does the agent fix a vulnerability on its own?

No. It reports the finding and proposes the update that remediates it, usually a plugin, theme or core update. Updates always stop for your approval, and the agent can take a backup first and rescan afterwards to confirm the finding is gone.

Other agents