AdLoop Cloud plugins

Google Ads, Reddit Ads and Google Analytics in your AI assistant, with a preview and your approval before every change.

This repository packages AdLoop Cloud as a plugin for Claude, Cursor, Gemini CLI, and ChatGPT and Codex. Each plugin does two things:

1. Connects the AdLoop Cloud MCP server at https://mcp.getadloop.com/mcp. It gives your assistant tools for Google Ads, Reddit Ads, Google Analytics 4, Search Console, Tag Manager and Merchant Center.
2. Adds AdLoop's orchestration rules. These rules tell the assistant which tool to use for which question and which checks to run before a change. They also cover how to read the data, for example why Ads clicks and GA4 sessions differ under cookie consent. The rules ship as a skill for Claude and Codex, as a rule for Cursor and as context for Gemini CLI.

The plugins contain no executable code. No hooks, no local servers, nothing that runs on your machine. The only network connection they make is to https://mcp.getadloop.com/mcp.

What you need

• A free AdLoop Cloud account: sign up at getadloop.com.
• Your Google account connected in the AdLoop Cloud dashboard. Connecting Reddit Ads is optional.

The first time your assistant uses AdLoop, it opens a browser window. Sign in to AdLoop Cloud there and approve the connection. You don't paste an API key anywhere.

Install

Claude (claude.ai, Cowork, Claude Code)

Once the plugin is listed, add AdLoop Cloud from Claude's plugin directory.

To try it from source in Claude Code:

git clone https://github.com/kLOsk/adloop-plugins
claude --plugin-dir ./adloop-plugins

Cursor

Once the plugin is listed, install AdLoop Cloud from the Cursor Marketplace.

To try it from source, clone this repository into ~/.cursor/plugins/local/adloop and restart Cursor.

Gemini CLI

gemini extensions install https://github.com/kLOsk/adloop-plugins

ChatGPT and Codex

Once the plugin is listed, add AdLoop Cloud from the plugin directory in ChatGPT or Codex.

Any other MCP client

Add https://mcp.getadloop.com/mcp as a custom connector or remote MCP server (Streamable HTTP). The server supports OAuth with dynamic client registration, so most clients only need the URL. For a client without OAuth support, create an API key under Settings → API keys in the dashboard and send it as a header:

Authorization: Bearer alc_YOUR_KEY

API keys can also be restricted to read-only access or to selected toolsets. See the connection guides for per-client instructions.

Safety model

AdLoop enforces these rules on its server, so they apply even to an assistant that ignores its instructions:

• Preview and approval. Every change starts as a preview. Nothing is applied until you approve it.
• Google validation test run. Before a change is applied for real, Google checks it in a test run that changes nothing. A change Google would reject never reaches your account.
• Budget and bid caps. Budgets above the maximum daily budget you set in the dashboard are refused, and so are bid raises beyond the per-change limit.
• New campaigns start paused. New campaigns and ads are created paused. You enable them after you've reviewed them.

Every change, including test runs, is recorded in the change log in your dashboard. More in the safety model docs.

What's in this repository

Keeping the rules in sync

The skill, the Cursor rule and GEMINI.md are generated from AdLoop's canonical rules file (.cursor/rules/adloop.mdc in kLOsk/adloop), rewritten for AdLoop Cloud. The generator and a manifest check live in that repository under scripts/plugins/, so this repository ships no code of its own: only manifests, the skill and assets.

Links

• Website: getadloop.com
• Documentation: docs.getadloop.com
• Privacy policy: getadloop.com/privacy
• Terms of service: getadloop.com/terms
• Support: hello@getadloop.com

License

MIT