Integration · Codex CLI × Google Ads

A Google Ads agent that runs in your Codex CLI

NotFair connects OpenAI's Codex CLI to your live Google Ads account through MCP. Diagnose campaigns, draft optimizations, and approve writes — all from the terminal.

Supported

Single `codex mcp add` command and you're running

Native MCP — no API key shims, no glue scripts

Approval-gated writes keep the agent on rails

Connect your Google Ads account

Authorize once at notfair.co — then open Codex and start asking real questions about your account.

Connect Google Ads

Setup

Connect Codex in three steps

Codex CLI ships with native MCP support. NotFair drops in as one of its MCP servers.

1. Register NotFair as an MCP server
From any shell where Codex CLI is installed.
```
codex mcp add NotFair-GoogleAds --url https://notfair.co/api/mcp/google_ads
```
2. Authorize Google Ads
Open notfair.co/connect, sign in, complete OAuth. Tokens are scoped to the accounts you authorize.
3. Start a session
Open Codex CLI in any directory and ask: "Diagnose my Google Ads account. What's wasting spend?"

Capabilities

What Codex can do with NotFair

Native MCP support

Codex CLI handles MCP servers as first-class citizens — no bridges or wrappers.

Live account reads

Performance, search terms, change history, landing-page status, conversion setup.

Approval-gated writes

Bids, negatives, budgets, ad copy, campaign state — all gated and logged.

Scriptable

Wrap NotFair workflows in shell scripts for recurring weekly audits.

Example workflows

Prompts that lead to real account work

Weekly account check

“Give me this week's Google Ads health report. What changed, what broke, what to fix this week. Be specific.”

A structured weekly report you can pipe into Slack or email.

Budget reallocation

“Across all my campaigns, find the ones that are budget-constrained AND have CPA below target. Propose a budget reallocation.”

Ranked list with current spend, lost-impression-share-to-budget, target CPA, and a proposed reallocation.

Ad copy refresh

“Find ad groups where every ad has been running over 90 days with declining CTR. Draft new responsive search ad headlines that I can review.”

Drafts new RSA assets per ad group; you approve before publishing.

Why NotFair for Codex CLI

Codex CLI is the cleanest agent surface OpenAI ships. NotFair is the Google Ads tool layer designed for exactly this use case — typed primitives, no GAQL-by-LLM, safe writes.

Typed tools beat raw GAQL — the agent gets a stable contract
Freshness metadata stops the agent from quoting yesterday's numbers
Approval-gated writes keep destructive actions on rails
Designed for terminal-native operators who manage spend themselves

FAQ

Common questions about using OpenAI Codex CLI with Google Ads through NotFair.

Does NotFair work with Codex CLI's auto-execute mode?

Read tools, yes. Write tools still pass through NotFair's approval gate even if Codex CLI is configured to auto-execute — this is intentional, because the gate protects your real ad spend.

Can I use NotFair in CI?

Read-only workflows yes. We don't recommend running write workflows in CI without a human approval step. NotFair will support team-level approval workflows for CI scenarios on the roadmap.

Is NotFair safe to use on a real Google Ads account?

Yes. NotFair separates read tools from write tools. Diagnostic queries run freely, but every write — bid change, negative keyword, ad copy, campaign state — is approval-gated and logged with full provenance so you can audit exactly what the agent did.

How much does NotFair cost?

Free during open beta. NotFair plans to keep a generous free tier for solo operators and a usage-priced plan for teams managing multiple accounts. See /pricing for current limits.

Is NotFair an official Google or Anthropic product?

No. NotFair is an independent product built on the Model Context Protocol (MCP) standard and the Google Ads API. MCP is Anthropic's open protocol for connecting AI clients to external tools; NotFair is one focused entirely on Google Ads.

Other integrations

Switching clients, or curious how this compares?