Build it yourself with Claude Code, or just use Roger.
Two ways to run AI-powered Google Ads management. One takes weeks of setup and ongoing maintenance. The other takes three minutes. Same intelligence, zero infrastructure.
14 hard steps, or 3 easy ones.
Same intelligence. Wildly different experience getting there.
Build it yourself
14 steps- 01 Set up Claude Code and your project Install Claude Code locally, create a new project, configure your Anthropic API key, scaffold a workspace, and pick a place to keep your prompts under version control. This is the easy part.
- 02 Create a Google Cloud project Every Google tool you want Claude Code to touch (Ads, GTM, Search Console, Merchant Center, GA4) needs OAuth credentials, and those credentials live in a Google Cloud project. Spin one up, enable the right APIs, generate the client ID and secret.
- 03 Install the MCP servers Find, install, and configure an MCP server for each Google product: Ads, Tag Manager, Search Console, Merchant Center, Analytics. One server per product. One OAuth flow per server. Read the configs, fix the inevitable token-refresh edge cases.
- 04 Get your Google Cloud project verified Without OAuth verification you're stuck on a 100-user test-mode quota, which is fine for one account but not for an agency. Submit your project to Google, add your logo, your privacy policy, your scopes. Add test users while you wait. Iterate when Google asks for changes.
- 05 Request a Google Ads API developer token No token, no read or write access on Google Ads at scale. Submit your use case to Google, justify your access tier, and wait roughly three days for approval. There is no shortcut. This is how Google gates the API.
- 06 Connect your MCC and sub-accounts If you manage clients under an MCC, this gets messy fast. Claude Code has no native concept of "which client am I in right now." You'll be manually scoping every conversation, every tool call. Cross-account confusion is the default state.
- 07 Realise the agent is generic At this point Claude Code is technically connected. The advice it gives is the kind of advice Google itself gives: apply the recommendations, trust the auto-bidding, broaden the match types. The model is excellent. It just hasn't been taught how a senior PPC strategist actually thinks.
- 08 Build the skills This is the real work. Roger ships with 20+ peer-reviewed skills: wasted-spend finder, negative keyword discovery, Quality Score recovery, Performance Max audit, Smart Bidding diagnostic, Recommendations decoder, funnel diagnostic, geo and device optimization, auction insights, Shopping feed health, and more. Each one is a battle-tested playbook. You'd be writing every single one yourself, then testing it on real accounts.
- 09 Inject business context The agent needs to know which conversion actions matter, what your margins are, who your audience is, what your brand voice sounds like. Without that, every recommendation is generic. You design and maintain that knowledge base, per client, by hand.
- 10 Build the guardrails Decide what the agent can do unattended and what must come back to a human. Drafts required for every change. New entities start paused. One website per ad group. No BROAD match on Manual CPC. Negatives stay PHRASE or EXACT. Caps on single-step budget increases. Caps on bulk operations. Hard daily spend ceilings. Untouchable campaigns. Lock down new campaigns. Force human review on routines. Now enforce all of that in code.
- 11 Schedule the routines Claude Code doesn't run by itself. You'll need cron jobs or a scheduler to run weekly spend reviews, daily search-term audits, monthly Quality Score checks. For each one: write the prompt, schedule it, capture the output, alert when something looks wrong, store the history so you can compare week over week.
- 12 Build a white-label reporting agent Agencies need branded reports for clients. That's a separate problem: pull data from Ads, GTM, Search Console, GA4 and Merchant Center, normalise it, format it, render it as a PDF, apply your agency's logo and colours, ship it. None of that is shipped by Claude Code out of the box.
- 13 Add multi-user access management If you're an agency with collaborators, each person needs their own scoped access: which clients they can see, which they can edit, who can apply changes vs draft-only. That's an auth layer with roles, audit logs, and per-account permissions. You design and run it.
- 14 Maintain it forever Google Ads doesn't sit still. New campaign types, new API versions, new policy rules, new bidding strategies, new Recommendations to ignore. Every change in Google's product is a change in your skills, your guardrails, your routines. This isn't a one-off setup project. It's a permanent engineering job.
Use Roger
3 steps- 1 Sign up to Roger Free, no card. Full access from day one.
- 2 Connect your accounts Google Ads MCC in one click. GTM, GA4, Search Console, Merchant Center: one click each. No developer token. No Google Cloud project.
- 3 Start optimizing 20+ peer-reviewed skills already loaded. Guardrails preconfigured. Routines scheduled. White-label reports included. Multi-user access ready.
What you build vs what you get.
Built by advertisers.
For advertisers.
Between us, we've spent hundreds of thousands of dollars on Google Ads, managing dozens of accounts and building several SaaS products along the way.
We know what good PPC management looks like because we've done it, at scale, with our own money. We've felt the frustration of wasted spend, the tedium of weekly audits, the risk of a budget that runs unchecked overnight.
When Claude Code came out, we saw the potential immediately: AI could finally manage Google Ads the way a senior PPC strategist would. We also saw the problem. Every agency we talked to wanted to build their own AI agent, and every one of them got stuck on the same wall: OAuth, MCP servers, developer tokens, skills, guardrails, routines, multi-user permissions. Months of engineering before a single ad got optimized.
So we built Roger. Everything we wished we had on day one. Now agencies can skip the infrastructure layer and start managing campaigns better, today.
Questions, answered.
Isn't Roger just a wrapper around Claude?
Roger is built on Claude. The model is the engine, not the car. The value is everything that sits around it: the 20+ skills, the guardrails, the connectors, the routines, the multi-user agency layer, the ongoing maintenance as Google Ads evolves. We're honest about this. If you only needed the model, Claude alone would be enough. But you also need the rest, and that's what Roger ships.
What if I already built my own Claude Code setup?
Then you already know how much of the work isn't the model. It's the skills, the guardrails, the routines, the reporting, the access control. You can keep your setup and use Roger as a complement on the accounts where speed matters. Or you can switch over once you compare the maintenance cost.
What data do you store?
Account metadata and the changes Roger has drafted or applied on your behalf. We do not copy your full Google Ads history into our database. You can disconnect any account in one click and we delete your data within 30 days.
Is my Google Ads data shared with Anthropic?
Claude is the model we use under the hood, served via Anthropic. Your prompts and the relevant context are sent to the model when you ask Roger to do something. Anthropic does not train on data sent through their API. Your campaign data is never used to train models. Yours or anyone else's.
Can I self-host Roger?
Not today. The hosted product is what we maintain, secure, and update every month. If you have a specific need for a private deployment (an agency with strict data residency requirements, for example), book a demo and we'll talk.
Pricing?
Free plan with full feature access. Team plan is per-seat. Standard at $20/seat/month with 1,000 credits. Premium at $100/seat/month with 5,000 credits. See the full breakdown on the pricing page.
Yes, you can do this yourself.
But you probably don't want to.
Claude Code is powerful. The MCP ecosystem is real. Google's Ads API is well-documented. Nothing about the DIY path is theoretical. It's been done, and it works.
But if you'd rather skip the infrastructure and start optimizing campaigns today, Roger is here. Same intelligence, zero infrastructure. Free to start.
