Adios
deploy an AI-built app

Codex app deployment

Deploy an app built with Codex to production.

Connect Codex to Adios over MCP and let the agent carry the project from a source-backed workspace to a verified preview. You keep visibility into the diff, runtime contract, logs, and final production decision.

Source to production

  1. 01Codex source
  2. 02Adios workspace
  3. 03Build + preview + logs
  4. 04Human production approval

Production stays separated from the agent's preparation work until you approve the release.

Source

Reviewable files and config

Preview

Healthy before production

Evidence

Build output and runtime logs

Control

Human approval gate

Deployable projects

Keep the generated app. Add the production contract.

Adios works from the project source and its normal framework commands. It does not require the app to be regenerated in a proprietary format.

  • Web apps generated or maintained with Codex
  • APIs, services, and background workers
  • Monorepo applications with explicit build paths
  • Projects that need databases, caches, queues, or secrets

Direct MCP connection

Give Codex a cloud it can operate.

Add the hosted Adios MCP endpoint to Codex and sign in through OAuth. Codex manages the connection while Adios scopes tool access to the authorized user and team.

Terminal
codex mcp add adios --url https://api.adios.dev/v1/mcp
codex mcp login adios

Example agent prompt

Ask for a verified release, not only a deploy command.

Run this from a Codex session after codex mcp login adios succeeds.

Prepare this app for production on Adios. Inspect the repository, identify its framework and runtime needs, write a reviewable adios.yaml, run the real production build, use logs to repair any failure, and return a healthy preview plus the exact changes that need my approval.

Reviewable configuration

The agent writes down how the app should run.

The exact manifest should match the repository. This example makes a conventional Node.js production build, port, and health endpoint explicit so the release can be reviewed and repeated.

Learn how adios.yaml works
adios.yaml
name: vibe-app
region: de

build_cmd: npm ci && npm run build
start_cmd: npm start

runtime:
  name: node@24
  port: 3000
  health_path: /api/health

Source-to-release workflow

Five checks between generated code and production.

  1. 01

    Inspect the project source

    Identify the framework, package manager, build output, start command, expected port, environment variables, and any stateful services before creating a release.

  2. 02

    Make the runtime contract visible

    Write or update adios.yaml so the build command, start command, runtime, port, health check, secrets, and managed resources can be reviewed beside the code.

  3. 03

    Build and open a preview

    Run the production build in an isolated workspace, start the application, and inspect a preview URL before treating the project as deployable.

  4. 04

    Read logs and repair failures

    Use build output, runtime logs, and the configured health path to diagnose failures. Apply the fix to source, then rebuild the same project rather than patching production by hand.

  5. 05

    Review and approve production

    Check the final source diff, configuration, preview, and health result. A person approves the production deployment after the agent has prepared a healthy release.

When the build fails

Give the agent evidence it can act on.

Codex can connect an error in the build or runtime logs to the current source and deployment manifest. After making a reviewable correction, it rebuilds and checks the configured health endpoint before requesting approval.

build · failed

↓ inspect output

source + adios.yaml · updated

↓ rebuild and health check

preview · healthy

↓ human review

production · ready for approval

FAQ

Deploying Codex apps with Adios.

Can Codex deploy through Adios directly?

Yes. Connect the hosted Adios MCP server to Codex, complete OAuth, and the agent can use scoped workspace, build, preview, log, and deployment tools. Production remains approval-gated.

Does Adios deploy only Next.js apps?

No. Adios supports explicit build and start commands for common frontend, full-stack, API, worker, and data-backed projects. The detected framework changes the runtime contract; the source-to-preview-to-production workflow stays consistent.

Can the AI agent deploy to production without approval?

No. The agent can prepare source, configuration, builds, previews, and diagnostics, but the production transition is approval-gated so a person can review the release first.

What happens when the production build fails?

The build output and runtime logs remain available to the workspace. The agent can use that evidence to update the source or configuration and run a new build before asking for production approval.

From source to healthy release

Let Codex prepare the deploy. Keep production approval yours.

Start with the source you already have, verify it in an Adios preview, and promote only the release you reviewed.

Start free trial