Codex + xCloud

Run Docker apps with Codex on xCloud

Running Docker apps with Codex means telling OpenAI's coding agent, from the terminal in your repository, to scan the Compose file, create the Docker site on xCloud after your approval, back it up before you change it and retry the deploy if it fails.

Skill: xcloud:deploy
Toolsets: sites, servers
Free with every xCloud account

Set up Codex All Codex guides

YouRun docker compose config, then scan the compose file on main with xCloud.

AgentLocal config resolves two services, api and cache. Only api publishes 3000.

git_compose-scanread-only

AgentxCloud's scan of main agrees: port 3000 from docker-compose.yml. Sending the dry run to the Frankfurt Docker server.

servers_sites_git_dockerdry run

AgentDry run resolved with no warnings. Create the site?

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 Run Docker apps 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

Run docker compose config here, scan the compose file on main with xCloud and tell me if the ports agree.

In practice

How Does Docker apps Work from Codex?

Codex sits in the repository and can run commands there, which gives it a check no dashboard has. If Docker is installed on your machine, it can run docker compose config against the file before anything leaves your laptop and read the resolved services and ports, then ask xCloud for git_compose-scan on the pushed branch and put the two side by side. The two answers should agree. If they do not, usually because of a Compose file in a subdirectory or an unpushed edit, Codex tells you which one it trusts and why. xCloud's scan is the one that counts, since that is what the server will run.

Once you have run codex mcp login xcloud, the exchange is a few lines. You name the Docker server, Codex runs git_detect and git_compose-scan, and reports the services, the published ports and the suggested primary port. It sends the Docker create request as a dry run and prints the resolved configuration: compose file name, port, address and warnings. You approve once, and it sends the same body with confirm set to true and an idempotency key. Then it polls the site status until the state is terminal, reads failed_steps and the SSL block and fetches the URL before it says the app is live. How often Codex pauses for you on its own commands is your approval setting; xCloud's confirmation on the create is separate and always applies.

For upkeep, Codex can take a backup with sites_docker_backup before you change an image tag, then read the result with sites_docker_backups. Private registries are the usual snag: the deploy runs docker compose pull and never docker login, so Codex proposes replacing the registry image with a build context, shows the diff, and commits it only when you agree. If a deploy still fails it reads sites_deploy-diagnosis, lists the settings you can correct, and retries on the same site. A restore is a dashboard step, and Codex says so instead of attempting it.

Codex specific: The xCloud tools stay unavailable until codex mcp login xcloud has completed, and a Docker run has no useful fallback without them, so type /mcp first and confirm xcloud lists its tools. On a machine with no browser, such as a CI runner that takes a nightly backup, set bearer_token_env_var in config.toml to the name of an environment variable holding a token with the mcp:invoke scope plus read:sites and write:sites (a backup is a write), and keep the token itself out of the file.

What xCloud does for docker apps

xCloud reads the repository, scans the Compose file for services and ports before anything is created, previews the site with a dry run and deploys it behind its own nginx once you approve. Docker sites can be backed up on demand and on a schedule, and a failed deploy is diagnosed and retried on the same site.

Pick a Docker server. The agent uses the server you name, or lists your servers and asks. A Docker deploy needs a Docker server; any other stack answers with an incompatible_server refusal, and an agentic server never takes a second site. WordPress is not supported on Docker servers.
Detect the repository. git_detect reads the repository root for a Dockerfile or one of the four Compose file names, and checks access and server compatibility. It creates nothing, and an access problem is never worked around by naming an app type by hand.
Scan the Compose file. git_compose-scan runs before the dry run. It lists the services and the host ports the file publishes, resolves which Compose file name it found, and suggests a primary port. The agent never guesses the port.
Dry run, then one approval. The agent sends the Docker create request with dry_run set to true and shows the resolved configuration: compose file, port, address and warnings. It asks once, naming the server and the URL, before sending the same body with confirm set to true and an idempotency key.
Poll and verify. A 202 means queued, not running. The agent polls the site status until it is terminal, reads failed_steps and the SSL block, and fetches the URL before it calls the app live. A 502 here usually means the Compose file publishes no port.
Back up and recover. The agent can take a Docker backup, read the backup list and settings, and add a note to a backup. If a deploy fails it reads the diagnosis, proposes a fix from the correctable fields and retries on the same site after your approval. Restoring a backup is a dashboard step.

Reference

Docker apps Settings and Limits on xCloud

The facts Codex works within when it runs Docker apps. Where a row names the dashboard, that step stays yours to take there.

Setting or limit	What applies
Supported inputs	A Dockerfile for a single container, or a Compose file for one or more services, on a Docker server. Go, Python, Rust and Java apps all deploy this way
Server stack	Docker servers only. The Docker deploy on any other stack, including an agentic server, answers 422 with the code incompatible_server
Which Compose file runs	docker-compose.yml at the repository root. A file named compose.yaml, compose.yml or docker-compose.yaml, or one in a subdirectory, deploys through the pinned Docker create with docker.compose_file set to the path the scan resolved
Public port	xCloud's nginx owns ports 80 and 443 and proxies to one host port of 1024 or higher on 127.0.0.1. xCloud runs your Compose file as-is and never rewrites its ports mapping
Port refusals	A port another site already holds is refused as port_unavailable. A port the file does not publish is refused as port_not_published, with the published ports listed
No published port	A Compose file that publishes no port is accepted and then answers 502, so the scan is the step that catches it
Private registries	The deploy runs docker compose pull and never docker login, so an image from a private registry cannot be pulled. Build the image from source instead
Environment and redeploys	Send environment values through env_file_content. A redeploy runs git reset --hard and git clean -df in the site directory, so uncommitted server-side changes are lost
What a Docker backup holds	Named volumes, eligible bind mounts, the Compose file and the app's .env, in an encrypted snapshot. The app is stopped while the backup captures and restarted afterwards
Backup tools	Take a backup, list backups, read the count and the settings, change the settings, add a note and delete a backup. Remote copies go to a storage provider you add in the dashboard
Restores	A restore replaces the app's data in place, so it stays a dashboard step: Site, Site Backup, Previous Backups, Restore Backup
Dot-prefixed paths	A 403 on a path that starts with a dot is the dot-path allowlist. It is fixed with docker.allowed_dot_paths in the deploy settings, without a rebuild

Rules Codex has to follow

The Compose file is scanned before the dry run, and the port always comes from the scan, never from a guess.
A file that binds ports 80 or 443, publishes no port or pulls from a private registry is not deployed as-is. The agent shows the rewrite, such as publishing 127.0.0.1:<high port>:<app port> and building the image from source, and the change goes into your repository.
Creating a site, deploying, redeploying and deleting a backup stop for your approval. A backup runs without a separate prompt but stops the app while it captures, so ask for one at a quiet time.
A failed Docker site is retried in place with corrections. It is never deleted and recreated, because it keeps its domain and port.
Restoring a backup, adding a backup storage provider and changing a live site's domain are dashboard steps; the agent gives you the path and the dashboard link xCloud returned.

Example prompts

What Can You Ask Codex to Do for Docker apps?

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

Run docker compose config here, scan the compose file on main with xCloud and tell me if the ports agree.

Prompt

Deploy this repo to the Frankfurt Docker server as a Compose app. Dry run, then wait for my yes.

Prompt

Take a Docker backup of notes.example.com, then replace the private registry image with a build context and show me the diff.

Prompt

Scan the docker-compose.yml in github.com/acme/api and tell me which services and ports xCloud would use.

Prompt

Deploy github.com/acme/shop to the Frankfurt Docker server as a Compose app. Scan the compose file first, show me the services and ports, then wait for my approval before creating anything.

Prompt

Deploy the main branch of github.com/acme/api to my Docker server as a Compose app on a staging hostname.

Prompt

My Compose app on notes.example.com answers 502. Check which port its compose file publishes and which port xCloud is using.

Prompt

Take a backup of the Docker app on notes.example.com now, before I upgrade it, and tell me when it is done.

Prompt

Show the backup settings and the last backup of every Docker app on this team, and list the ones with no schedule.

Prompt

The last deploy of the API site failed. Diagnose it, show me the settings I can correct, fix the build command and retry on the same site.

Codex and Docker apps: Frequently Asked Questions

What people ask before they let Codex run Docker apps through xCloud.

Can Codex test my Compose file locally before deploying to xCloud?

Yes, if Docker is installed on your machine. Codex can run docker compose config and read the result, but xCloud's own git_compose-scan of the pushed branch decides the port. Treat the local run as a pre-check.

What does Codex do when my Compose file pulls from a private registry?

It stops and explains, because xCloud's deploy runs docker compose pull and never docker login. Codex can propose a build context in place of the image, show you the diff and commit it after you agree, then scan and deploy again.

Which servers can run a Docker app on xCloud?

Docker servers. Node.js, PHP and static builds use Nginx or OpenLiteSpeed servers, and the Docker deploy on any other stack is refused with an incompatible_server code. An OpenClaw, Hermes, Paperclip or DeepSeek Harness server hosts only the site created at provisioning, so it takes no Docker app.

Why does the agent scan the Compose file before deploying?

xCloud proxies to the host port your file publishes and never rewrites it. The scan lists the services and ports, so the agent picks a port that is free on the server and published by the file instead of guessing.

Why does my Compose app answer 502 after it deployed?

The most common cause is a Compose file that publishes no port: xCloud accepts it, then its nginx has nothing to proxy to. Publish the app on 127.0.0.1 with a high port, such as 127.0.0.1:8080:8080, and redeploy.

Can an agent deploy an image from a private registry?

Not as-is. The deploy runs docker compose pull and never docker login, so a private image cannot be pulled. Build the image from source in the Compose file, or deploy a repository that does.

Can an agent back up and restore a Docker app?

It can take a backup, read the backup list and settings, change the settings, add a note and delete a backup. Restoring replaces the app's data in place, so you do that in the dashboard under Site, Site Backup, Previous Backups.

Can I run WordPress as a Docker app on xCloud?

No. Creating a WordPress site on a Docker server is refused because WordPress is not supported on Docker servers. Use an Nginx or OpenLiteSpeed server for WordPress.

Other agents