Adrex AI

Adrex AI

Open-source MCP server for Google Ads & Meta Ads

Manage your ad campaigns in plain English from Claude Code, Cursor, Codex, Gemini, or any MCP client.

[npm version] [npm downloads] [MIT license]

Quick Start | 64 Tools | Platform | Discord

What is Adrex AI?

Adrex AI is an open-source MCP (Model Context Protocol) server that connects AI assistants to your Google Ads and Meta Ads accounts. Instead of navigating complex dashboards, you manage campaigns, reporting, keywords, and targeting through natural language — from Claude, Cursor, Codex, or any MCP client.

Platforms: Google Ads and Meta Ads (Facebook & Instagram) today — TikTok and LinkedIn coming soon.

You: "Show me my Google Ads performance for the last 7 days"Claude: [calls google_ads_campaign_performance] → table with spend, clicks, CTR, ROAS
You: "Pause the campaign with the lowest ROAS"Claude: [calls google_ads_pause_campaign] → campaign paused, no more spend
You: "Create a Meta campaign for our summer sale, $50/day budget"Claude: [calls meta_ads_create_campaign] → campaign created PAUSED for your review

You connect your ad accounts once in the Adrex dashboard (secure OAuth — your tokens stay on our servers, never in the MCP client), then drive everything from your AI assistant. There are two ways to connect (see Quick Start): a remote server with browser sign-in — no API key, or this npx server with an API key.

Safety First

These tools operate on real ad accounts that spend real money:

• All campaigns and ads are created PAUSED — no spend until you explicitly resume
• Destructive and spend-starting actions return clear warnings
• Read operations (metrics, lists) run freely without side effects

Quick Start

First, sign up at adrex.ai and connect your Google/Meta ad accounts in Settings. Then pick one of the two ways to connect.

Option A — Remote server (recommended, no API key)

Connect over HTTP and authorize in your browser — nothing to copy or paste. Your client stores its own secure token. In the Adrex dashboard, Settings → Connect to your AI (MCP) shows a one-command connect with your server URL. It looks like:

Claude Code:

claude mcp add --transport http adrex-ai https://<your-adrex-server>/mcp

Claude Desktop / Cursor / VS Code / other clients (JSON) — or use the one-click Add to Cursor / Add to VS Code buttons in Settings:

{  "mcpServers": {    "adrex-ai": { "type": "http", "url": "https://<your-adrex-server>/mcp" }  }}

After adding, your client opens a browser tab — sign in and click Authorize. The consent screen shows which Adrex account you're authorizing as; if a different account is already signed in on that browser, use Not you? Switch before approving.

You can see and revoke every connected app in Settings → Connect to your AI → Connected AI apps.

Connected the wrong account? Every tool will report "Not connected" even though your ad accounts are linked. Revoke the app in Settings, then re-authorize. Note that re-running claude mcp add does nothing once the server is configured (already exists in local config) — instead run /mcp in Claude Code, pick adrex-ai, and clear authentication.

Option B — npx server with an API key

Prefer the classic stdio server? Generate an API key in Settings, then add it below. Adrex ships a standard stdio MCP server, so it works with any MCP client — every config points at npx -y adrex-ai with your ADREX_API_KEY.

Claude Code:

claude mcp add -s user adrex-ai -e ADREX_API_KEY=your-api-key -- npx -y adrex-ai

Gemini CLI:

gemini mcp add adrex-ai -e ADREX_API_KEY=your-api-key -- npx -y adrex-ai

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json), Cursor (~/.cursor/mcp.json global or .cursor/mcp.json per-project), Gemini CLI (~/.gemini/settings.json), or any client using the JSON mcpServers format:

{  "mcpServers": {    "adrex-ai": {      "command": "npx",      "args": ["-y", "adrex-ai"],      "env": {        "ADREX_API_KEY": "your-api-key"      }    }  }}

Codex (~/.codex/config.toml) — note Codex uses TOML, not JSON:

[mcp_servers.adrex-ai]command = "npx"args = ["-y", "adrex-ai"]env = { ADREX_API_KEY = "your-api-key" }

Any other MCP client (Cline, Windsurf, Zed, Continue, …): point it at command npx -y adrex-ai with env ADREX_API_KEY set to your key.

That's it — no SDK setup, no developer tokens, no OAuth juggling. The server forwards each request to the Adrex backend, which uses the ad-account credentials you connected in the dashboard.

Behind a corporate proxy? v1.0.4+ automatically routes through HTTP_PROXY / HTTPS_PROXY.

Tools

Google Ads (35 tools)

Campaigns

Ad Groups

Ads

Keywords

Reporting

Analysis

Meta Ads (26 tools)

Campaigns

Ad Sets

Ads

Targeting & Audiences

Reporting

Cross-Platform (3 tools)

Use Cases

Performance Review

"How are my Google Ads campaigns performing this month? Show me the ones with ROAS below 2x."

Campaign Management

"Create a new Meta traffic campaign called 'Summer Sale 2026' with a $75/day budget targeting ages 25-45 in the US."

Keyword Optimization

"Show me keywords in my 'Brand Terms' ad group. Pause any with quality score below 5."

Budget Control

"Increase the budget on my top-performing campaign by 20% and pause the one that's losing money."

Want More?

The MCP server gives you full campaign management from chat. The Adrex AI Platform adds:

• AI Campaign Agent — autonomous optimization with guardrails and rollback
• Creative Studio — AI-generated ad copy and images
• A/B Testing — native platform experiments
• Competitor Intelligence — track competitor ads across Google and Meta
• Budget Rules Engine — automated rules with natural language parsing
• Visual Dashboard — interactive charts, breakdowns, and time-series

Links

• Adrex AI Platform
• Documentation
• Discord Community

License

MIT License — see LICENSE for details.

Built by the team behind Adrex AI