Connect your agent

You connect Ploa to your agent once. On your computer that's three commands; for an agent on the web, one address you add and approve. After that you talk to the agent, and it does the store work.

The short way

On the computer where you work, in the app's folder (or its repository):

curl -fsSL https://ploa.dev/install.sh | sh
ploa login
ploa mcp install

1. Install

   The first line installs Ploa's command line tool, ploa, which also runs Ploa's MCP server for your agent. On Windows, in PowerShell: irm https://ploa.dev/install.ps1 | iex (Working from Windows).

2. Sign in

   ploa login opens Ploa in the browser to confirm a code. There you choose how far your agents may go (below).

3. Add Ploa to your agents

   ploa mcp install adds Ploa to Claude Code, and to Cursor, VS Code (Copilot) and Codex when they're on this computer. Restart the agent, or reload its MCP servers.

Then ask the agent:

What you need

• A Ploa account. Sign up at ploa.dev/join, or from ploa login. Your stores don't need to be connected first: the agent walks you through connecting them.
• Node 24 or newer on the computer, for ploa and the server on your computer. Check with node --version.
• For builds on this computer: a Mac with Xcode for iOS, JDK 17 or newer and the Android SDK for Android. Not needed when the builds run on GitHub or in your own Google Cloud. ploa doctor says what this computer has.

Your agent, the shortest way

Install the CLI

The MCP server on your computer is part of the ploa command line tool. Choose one way:

curl -fsSL https://ploa.dev/install.sh | sh

irm https://ploa.dev/install.ps1 | iex

npm install -g @ploa/cli

npx -y @ploa/cli login
npx -y @ploa/cli mcp install

Sign in

The server works as you, with the access you choose. Sign in once on the computer:

ploa login

ploa login shows a code and opens Ploa in the browser (--no-browser only prints the link). Check that the code is the one in your terminal, then choose what the CLI and your agents on this computer may do:

An agent can only ask for what its level allows, so for an agent that takes the app all the way to the stores, pick Submit for review or Production. You can change it later in Settings › Tokens. Every level, and what waits for you: Access and approvals.

The sign-in is kept in ~/.config/ploa/config.json, readable only by you, never in your project. No browser on the machine (a server, CI)? Make a token on the dashboard in Settings › Tokens and set it as PLOA_TOKEN.

One command for your agents

In the app's folder (or its repository):

ploa mcp install

It adds Ploa to each agent's MCP settings and leaves the agents' other servers as they are. Claude Code always gets it; the others when they're on this computer:

The project's top folder is the git repository's top, or the app's folder when there's no repository. Only some of the agents? Name them (claude, cursor, vscode, codex):

ploa mcp install --agent claude,cursor

For each agent it says added, updated (an older Ploa entry replaced), unchanged or skipped, with the file. Then restart the agent, or reload its MCP servers.

The files hold no secrets, but they hold this computer's full paths to Node and to ploa (agents opened from the Dock don't get your shell's PATH), so everyone on your team runs ploa mcp install on their own computer. Installed from npm, the files run npx -y @ploa/cli mcp instead.

Without installing: npx

Any agent that can start a command can run the server from npm:

npx -y @ploa/cli mcp

Sign in once on the computer with npx -y @ploa/cli login, or give the server a token. What you can set:

Settings by hand, agent by agent

These use npx, so they work on any computer with Node 24. If you installed the CLI, "command": "ploa", "args": ["mcp"] works the same, as long as the agent sees your PATH.

claude mcp add ploa --scope project -- npx -y @ploa/cli mcp

{
  "mcpServers": {
    "ploa": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@ploa/cli", "mcp"]
    }
  }
}

{
  "mcpServers": {
    "ploa": {
      "command": "npx",
      "args": ["-y", "@ploa/cli", "mcp"]
    }
  }
}

{
  "servers": {
    "ploa": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@ploa/cli", "mcp"]
    }
  }
}

codex mcp add ploa -- npx -y @ploa/cli mcp

[mcp_servers.ploa]
command = "npx"
args = ["-y", "@ploa/cli", "mcp"]

{
  "mcpServers": {
    "ploa": {
      "command": "npx",
      "args": ["-y", "@ploa/cli", "mcp", "--dir", "/path/to/your/app"]
    }
  }
}

{
  "mcpServers": {
    "ploa": {
      "command": "npx",
      "args": ["-y", "@ploa/cli", "mcp", "--dir", "/path/to/your/app"]
    }
  }
}

On the web: ploa.dev/mcp

For agents that can't run a command on your computer, like Claude on the web, Ploa runs the server itself at:

https://ploa.dev/mcp

1. Add it to the agent

   In Claude: Settings › Connectors › Add custom connector, and paste the address. Other agents call it a remote MCP server or a connector.

2. Approve it on Ploa

   The agent sends you to ploa.dev. Sign in if you need to, check the agent's name and where it returns to, and choose what it may do: Read only; Drafts and builds (store page drafts and builds on GitHub); or Drafts, builds and requests, which also asks you for App Review, production, prices, testers and replies to reviews. Then Allow.

3. Start working

   With one app in the workspace, the agent works on it. With more, it names one by its ID (ploa_apps lists them).

What's different on the web:

• There's no project folder, so builds run on GitHub from what's pushed (set up once, see GitHub), and the server's code comes from the repository too.
• What needs your computer isn't offered: the setup guide and linking a folder, preflight, builds on this computer, simulator screenshots, privacy hints from the code, the Android upload key, the steps-only-you-can-do list, and anything uploaded from your disk (screenshots, videos, images, documents). Use the server on your computer for those. Every MCP tool marks the tools that are on this computer only.
• An agent on the web asks; it doesn't act past drafts and builds. With Drafts, builds and requests it asks you for App Review, production, prices, testers, products and replies, and each waits for your approval in Ploa. It never changes the stores directly (screenshots, TestFlight groups, sending the store page) and can't be autonomous: those are done on the dashboard, or by the agent on your computer.
• It can't be made autonomous.
• The approval becomes an ordinary Ploa token, named after the agent with "(web)", at the level you chose. It's listed in Settings › Tokens, where you can revoke it at any time, and the workspace's owners are told when it's made.

In CI and GitHub Actions

An agent running in CI uses the same server, with a token from Settings › Tokens kept in the CI's secrets and passed as PLOA_TOKEN. In GitHub Actions:

env:
  PLOA_TOKEN: ${{ secrets.PLOA_TOKEN }}

Give CI a token with the lowest level it needs. Builds Ploa runs on GitHub don't need a token in the repository at all: Ploa's own workflow signs in as the run.

Check that it works

1. Restart the agent and open its list of MCP servers: ploa should be connected, with its tools.
2. Ask: Which Ploa account and access do you have? The agent calls ploa_access and answers with your name, level and how it works (autonomous or approvals).
3. Ask: Where does this app stand in the stores? The agent calls ploa_setup_guide for a new app, or ploa_shipping_report.

Something off? See Troubleshooting.

Update and remove

• Update: ploa update (install script), npm install -g @ploa/cli@latest (npm); npx runs the newest by itself. Then restart the agent: the Ploa tools it already started keep running the old version until then, and say so.
• Remove from an agent: delete the ploa entry from the agent's MCP settings file.
• Sign out this computer: ploa logout signs out and revokes this computer's token.
• Revoke a web agent: Settings › Tokens on the dashboard.