Agent plugin

The Adside plugin teaches an AI agent on your computer how to work in Adside: when to use the command line and when to use the MCP, how to log in, how to find the right command, and what to ask you before anything runs. It works in Claude Code and Codex. It contains:

Three skills: adside (setup, login, finding a command, what needs your OK), adside-media (image, video and audio generation, priced before anything is paid) and adside-assets (uploading, downloading and organizing your asset library).
The Adside MCP server, https://api.adside.ai/mcp, for changes to your ad platforms.

The plugin does not include the command line. Install it first, or let the agent ask you to.

Install

Claude Code
Codex

claude plugin install adside --marketplace AgenticAdvertising/adside-plugin

Then type /mcp in Claude Code and sign in to Adside once.If you added the Adside MCP before with claude mcp add, remove that copy so your agent does not see every tool twice:

claude mcp remove adside

Updates are not automatic. To get the latest version:

claude plugin marketplace update adside

claude plugin update adside@adside

codex plugin marketplace add AgenticAdvertising/adside-plugin

codex plugin add adside@adside

Codex signs in to the Adside MCP the first time it uses it. If you added the Adside MCP before with codex mcp add, remove that copy.By default Codex runs commands in a sandbox with no network, so each adside command asks for your approval. To run some without asking, see below.

Let read commands run without asking

Both agents ask before each adside command until you allow it. Allow the commands you trust, one by one.

Claude Code
Codex

Add them to permissions.allow in ~/.claude/settings.json, or in .claude/settings.json for one project:

{
  "permissions": {
    "allow": [
      "Bash(adside whoami)",
      "Bash(adside cli search *)",
      "Bash(adside reporting performance query *)"
    ]
  }
}

If Claude Code runs in its sandbox, allow api.adside.ai and app.adside.ai in its network settings.

Add rules to ~/.codex/rules/default.rules, then restart Codex:

prefix_rule(pattern=["adside", "whoami"], decision="allow")
prefix_rule(pattern=["adside", "cli", "search"], decision="allow")

A command allowed this way runs outside the sandbox, with network and access to your login.

Never allow adside as a whole (Bash(adside *) or pattern=["adside"]): that would let the agent change things and spend media credit without asking you.

What the agent may do

The command line and the API make a change on the first call. There is no confirmation step, so the skills tell the agent to ask you before anything that creates, changes, deletes, sends or spends.
On ad platforms, the command line only reads, apart from posting to your Instagram account. Changes to campaigns, ad sets, ads, budgets and audiences go through the MCP, which asks for a confirmation first.
Media generation costs credit. The agent prices a job first, tells you the price, and caps each request with --max-price-usd.
The agent acts as you, with your permissions. For an agent that runs without you, in CI or in the cloud, use a personal API key, read only unless the job must change things.

Try asking

“Which of my Meta campaigns spent the most last week?”
“Upload the images in this folder to my asset library and tag them spring-launch.”
“Make three square product images from this photo. Tell me the price first.”

CLI

Command reference

Install

Let read commands run without asking

What the agent may do

Try asking

​Install

​Let read commands run without asking

​What the agent may do

​Try asking

Install

Let read commands run without asking

What the agent may do

Try asking