A proposal to adopt Anthropic's Claude as our AI infrastructure layer — starting with MCP-powered ad account automation across Google, Meta, and beyond.
The marketing industry is at an inflection point. Agencies that can connect AI directly to live client data will deliver faster optimisations, sharper reporting, and measurably better outcomes than those still exporting CSVs.
Claude's Model Context Protocol (MCP) makes this possible. It lets us build secure, authenticated connections between Claude and ad platforms — Google Ads, Meta, LinkedIn, and others — so our team can interrogate real account data in natural language and surface insights instantly.
This proposal recommends a Claude Team plan licence as the foundation for this capability, with a phased build starting in Performance.
Bottom line: A Claude Team licence at ~$20–100 per seat per month gives Rogerwilco the ability to build proprietary MCP tooling that replaces hours of manual reporting, accelerates optimisation cycles, and becomes a genuine agency differentiator — one competitors cannot easily replicate.
MCP (Model Context Protocol) is an open standard that lets Claude connect securely to external tools, APIs, and data sources. Think of it as giving Claude a live, authenticated window into our clients' ad platforms — rather than Claude working from exported snapshots or our descriptions of the data.
With custom-built ad account MCPs, Claude can:
Ask Claude "which campaigns are underperforming vs target ROAS this week?" and get a real answer from the account — not a report you export later.
Claude monitors accounts and proactively surfaces budget pacing issues, audience fatigue, or bid anomalies the moment they appear.
Generate structured performance commentary and insights directly from live account data — ready to drop into client decks or dashboards.
Compare performance across Google, Meta, and LinkedIn in a single query. No toggling between platforms or reconciling different export formats.
Analyse which copy variants are winning, identify patterns, and generate new ad copy hypotheses grounded in actual performance data.
Each MCP connection is scoped and authenticated per client — so there is no risk of data bleed between accounts.
Anthropic's Team plan supports mixed seat types, which maps well to our needs. Developers and power users building MCPs need Premium seats (which include Claude Code). The broader team consuming the output needs Standard seats.
Minimum 5 seats total. Mix and match freely — recommended starting point: 3 Premium + 5 Standard seats, scaling as MCP adoption spreads across the agency.
| Value driver | Estimated hours saved / month | At R650/hr blended rate |
|---|---|---|
| Manual performance reporting across accounts | 20 hrs | R13,000 |
| Ad copy analysis & variant generation | 12 hrs | R7,800 |
| Cross-platform data reconciliation | 8 hrs | R5,200 |
| Campaign audits & optimisation flagging | 10 hrs | R6,500 |
| Estimated monthly value | ~50 hrs | ~R32,500 / month |
At current USD/ZAR rates, the $400/month licence is approximately R7,200/month — a conservative 4.5× return on effort freed up from recurring manual tasks alone, before accounting for improved client outcomes or new capability positioned to clients.
Procure Team plan with 3 Premium + 5 Standard seats. Assign Premium seats to Werner + 2 developers. Set up central billing under Rogerwilco admin.
Use Claude Code to build the first MCP server connecting Claude to the Google Ads API. Scope: campaign performance queries, budget pacing, anomaly detection. Deploy for one pilot client account.
Extend to Meta Ads API. Onboard account managers to Standard seats. Build shared Claude Projects for each client with MCP context pre-loaded.
Expand to LinkedIn, GA4, and Search Console MCPs. Evaluate whether this becomes a productised client offering — "AI-powered account intelligence" as an Rogerwilco differentiator.
This is a low-risk, high-leverage investment. The $4,800/year entry cost is recoverable within the first month based on time savings alone. The strategic upside — building proprietary AI tooling that connects directly to client ad accounts — is material and durable.
No other tool in our current stack offers the combination of frontier AI capability + direct platform connectivity + the ability to build and own our own MCP layer. This is infrastructure, not just software.
I'm ready to proceed with procurement immediately upon approval. The first Google Ads MCP can be live within four weeks.