Turn Hermes Into Your Marketing Engine

Connect NotFair to Claude, OpenClaw, Codex, Cursor, or Hermes and operate Google Ads or Meta Ads from a chat - diagnose, draft fixes, and approve every write.

1

Add the MCP server

Run this in your terminal to register NotFair with Hermes:
hermes mcp add NotFair-GoogleAds --url https://notfair.co/api/mcp/google_ads
2

Test the connection

Ask Hermes to verify the MCP server and complete the sign-in flow if prompted:
hermes mcp test NotFair-GoogleAds
3

Start asking

Once the test succeeds, use Hermes with NotFair. Try this prompt:
Audit my Google Ads account and rank fixes by impact.

A Complete Marketing Engine Inside Your Agent

Diagnostics, recommended fixes, writes, and undo — all driven by one MCP connection per ad account.

Why did our cost-per-lead spike this week?
!FNotFair

CPL up 42% WoW. Three campaigns drove it:

  • Brand-Search+$1,820 spend - 0 conversions
  • PMax-US+$940 - CPA $38 -> $89
  • LeadGen-Q4search terms drifted to "free"

Want me to draft 14 negatives + pause Brand-Search?

Diagnose

Live context. Not yesterday's report.

Your agent reads the same numbers you would open in the platform UI: cost, conversions, impression share, learning phase, frequency, quality score. No CSV exports, no stale dashboards.

Find wasted spend and rank fixes by impact.
!FNotFair

Checked spend, search terms, quality scores, and assets. Top fixes by spend at risk:

  1. 01Loose-match keywords burning $2.4k/wk$2,400/wk
  2. 02Missing negatives on LeadGen-Q4$1,180/wk
  3. 03PMax with no asset coverage$640/wk
  4. 043 ads disapproved (silent)$310/wk

Total recoverable: $4,530/wk. Open the fix list?

Prioritize

Fix lists, not reports.

One prompt fans out the reads, scores findings by spend at risk, and turns the result into prioritized actions your agent can draft for approval.

Pause everything below 1.5 ROAS in the last 14 days.
!FNotFair

Found 6 ad sets matching. Diff before I apply:

Ad set14d spendROAS
Retarget-Cart-V3$1,2400.8
Cold-Lookalike-IG$8901.1
Broad-Interest-FB$6101.3
Influencer-Test-A$4200.4
awaiting approval
Operate

Pause, shift, rename, rewrite - from chat.

Not a passive reporting tool. Every entity an account manager touches is exposed as a write tool, with a diff-and-approve gate before anything hits the ad platform.

Undo what you did at 14:32.
!FNotFair

Found change chg_8f21 - paused Brand-Search at 14:32 today.

BeforePAUSED
After undoENABLED

Reverted. Anything else to roll back?

1 change reverted
Approve & undo

Every write is reviewed. Every write is reversible.

Reads are unrestricted. Writes show you a diff before they land, and NotFair logs every change to its own history - one call rolls anything back.

Built for the people spending the budget.

SMB founders

Self-doers running their own Google or Meta spend. Replace the agency review call with a Claude conversation.

Performance agencies

Onboard a new account, audit it, and ship the fix list before the kickoff call ends.

E-commerce ops

Daily search-term sweeps, negative-list maintenance, and budget reshuffles run from a single agent prompt.

Lead-gen teams

Match search terms to MQL quality, tighten match types, and route budget toward the geos and keywords that close.

Solo consultants

Manage ten accounts the way you used to manage one. One MCP endpoint per client, one chat per question.

Wire it up.
Operate from chat.

Connect Meta Ads

Free · OAuth 2.0 · No credit card

FAQ — NotFair MCP

Common questions about authenticating, configuring, and operating the NotFair MCP servers.

What is NotFair MCP?

Two hosted Model Context Protocol servers — one for Google Ads, one for Meta Ads — that expose your accounts to MCP-compatible AI clients. Reads stream live campaign data; writes are proposed in chat and require explicit approval before they hit the ad platform.

Which AI clients can I use?

Anything that speaks the MCP Streamable HTTP transport: Claude.ai (Web, Desktop, Cowork), Claude Code, OpenAI Codex CLI, Hermes Agent, OpenClaw, Cursor, Cline, and custom MCP clients. The server URL stays the same — only the client-side config differs.

Do I need to self-host anything?

No. Both servers are hosted by NotFair at https://notfair.co/api/mcp/google_ads and https://notfair.co/api/mcp/meta_ads. You connect the underlying ad account once via OAuth and point your client at the URL.

How does authentication work?

OAuth 2.0 with PKCE is the default — Claude.ai and Codex run it automatically. For clients that don't support OAuth, you can use a Bearer token via the Authorization header. Generate either at notfair.co/connect or notfair.co/connect/meta-ads.

Can the AI write to my ad accounts?

Only with explicit approval. Write tools propose changes, the client surfaces the diff, and you confirm before anything hits the Google Ads or Meta Marketing API. Read access is unrestricted; every write is gated.

What does it cost?

Connecting and running audits is free with no credit card. Paid plans unlock higher usage limits and team features.

Where do I find platform-specific configs?

The /google-ads-mcp and /meta-ads-mcp pages have generic JSON snippets that work in any MCP client. The per-client setup guides walk through Claude.ai, Claude Code, and Codex specifically.