Skip to content

Coding agents

Let your coding agent run your servers.

Claude Code, Codex or Cursor adds the server, ships the repo and sets up the database. Every deploy, restart or delete your agent asks for waits for you to approve.

your coding agent · ~/shop
  1. you › Add Postgres and connect web to it.
  2. $ ployz service add postgres --image postgres:18
  3. Staged new Service postgres in shop/production (revision 13).
  4. $ ployz volume add pgdata --mount postgres:/var/lib/postgresql
  5. $ openssl rand -hex 24 | ployz set postgres.env.POSTGRES_PASSWORD --secret
  6. $ ployz set 'web.env.DATABASE_URL=postgresql://postgres:${{ postgres.POSTGRES_PASSWORD }}@postgres.internal:5432/postgres'
  7. $ ployz diff
  8. postgres (create) · pgdata (create) · web (update) next: ployz deploy --expect-version 16:5:c81e
  9. agent › Staged, not deployed. It's on your canvas. Deploy?
  10. you › Yes.
  11. $ ployz deploy --expect-version 16:5:c81e --message "Add Postgres"
  12. Deployment #18 of shop/production: applied postgres: deployed web: deployed

ployz.dev · shop / production

web
web.shop.up.ployz.app
Online
postgres
Online
pgdata
Deployment #18 · applied

“Add Postgres and connect web to it.” The new service lands on your canvas, staged, with Deploy waiting.

Risky actions wait for you

Every deploy, restart or delete your agent asks for waits for you to approve. No setting turns it off.

Less context to read

ployz mcp gives the agent 55 tools, not one per endpoint. Every command takes --json, and a refused call names the exact call to retry.

Secrets stay sealed

Values come in on stdin with --secret, so they never appear in the agent's transcript.

Staged first. Deployed when you say.

The agent edits the same draft you edit in the dashboard. Have it run ployz deploy --expect-version, and it ships only the diff you read.

The canvas with a new service marked New and Not deployed, and the bottom bar reading Apply 1 change with Details and Deploy buttons

One line teaches it Ployz.

Paste Read ployz.dev/agents.md and deploy this repo to my server. into Claude Code, Codex or another coding agent. It sets Ployz up, then deploys the repo you're in to the server you rent. These are the steps it runs:

your laptop

$ curl -fsSL https://ployz.sh | sh

Installs the ployz CLI on Linux or macOS.

$ curl -fsSL https://ployz.sh/skill.md -o ~/.agents/skills/ployz/SKILL.md

Adds the Ployz skill, for Claude Code too.

$ ployz login

Opens your browser. Click Approve and the terminal is signed in.

$ claude mcp add -s user ployz -- ployz mcp

Gives the agent the Ployz tools.

Compared with the platforms.

What you needVercel, Railway, Render, Heroku, Fly.ioPloyz on your server
How agents connectAn MCP serverThe ployz CLI through their shell, or ployz mcp
Agent setupMCP config, skills or pluginsOne pasted line installs a skill and ployz mcp
Review before it goes liveVaries. Vercel deploys previews unless --prodStaged changes, shipped with --expect-version
CredentialsOften your whole accountOne organization, 90 days, revocable at once
Where it runsTheir compute, billed by usage or planServers you rent

Good to know.

  • An agent connected through ployz mcp can't answer an approval. You approve from your terminal, Ployz Cloud or the dashboard's agent panel. An agent in your own terminal holds your sign-in and can approve, so keep ployz on its ask list there.
  • ployz deploy runs at once. Put it on your agent's ask list if you want to read every diff first.
  • The dashboard has its own agent. It answers questions about your projects and can restart, start or stop services, with an approval for anything destructive.
  • What an agent deploys gets the same monitoring, searchable logs and a restore point for every database.
  • The public cases of agents wiping production, and the four layers that stop it, are in how to stop AI agents from deleting your production database.

Questions

Can Claude Code deploy my app to my own server?

Yes. Paste “Read ployz.dev/agents.md and deploy this repo to my server” into your agent. It installs the CLI, signs you in and deploys. Codex, Cursor and other agents with a shell work the same way.

What waits for my approval?

Through ployz mcp, every deploy or redeploy the agent asks for waits for you, and so do stopping or restarting a service and removing a service, a volume, a domain or a server. The agent can stage changes and read logs without asking, its token can't approve, and nobody approves their own request. No setting turns it off.

Does Ployz have an MCP server?

Yes. ployz mcp gives your coding agent 55 tools, the same ones the agent in the dashboard uses. Agents with a shell can also use the CLI, with --json output and the Ployz skill.

Will my agent deploy without asking me?

Not through ployz mcp: each deploy it asks for waits for your approval. An agent running the ployz CLI in your own terminal acts with your sign-in, so there ployz deploy runs at once. Put it on your agent's ask list and deploy with --expect-version.

How do I give an agent access in CI?

Run ployz token new with a name, store it as PLOYZ_TOKEN, and pass --project. Tokens expire after 90 days by default, and ployz token rm revokes one.

Does it work with Coolify or Dokploy?

Those are separate tools with their own MCP servers. Ployz is its own platform, built for coding agents on your own servers, and the agent drives it with the ployz CLI or ployz mcp.

What does it cost?

Nothing, on any number of servers, with unlimited projects, previews and custom domains. You pay your provider for the servers.

Your servers. The whole platform. Free on your own servers.

Deploy your app

Facts checked 2026-10-18.