Every week a new MCP lands in the marketing ecosystem. First came web search. Then CRM. Now Google Ads and Meta both have official MCP servers — and the question I keep getting from founders and growth leads is some version of: does this mean I can fire my paid media agency?
The short answer is: maybe. The honest answer is: it depends on exactly what you're paying that agency to do, and whether you understand the difference between access and judgment.
Here's my view, built from direct experience running paid campaigns across 13 African markets and now building an AI-native growth stack that plugs into the same APIs these MCPs are wrapping.
What the MCPs actually do.
A Model Context Protocol server is, at its core, a structured bridge between an LLM and an external platform's API. The Google Ads MCP exposes campaign creation, budget management, keyword planning, bid strategy adjustment, and performance reporting to any MCP-compatible AI agent. The Meta equivalent does the same for campaigns, ad sets, creative management, audiences, and attribution reads.
What this means in practice: you can instruct Claude or GPT-4o to "review last week's campaigns, identify the three worst-performing ad sets by CPA, and pause them" and it will actually execute those operations. Not draft them. Execute them. The AI has write access to your ad accounts.
That is genuinely powerful. For the right use cases.
The pros: where MCPs deliver real leverage.
- Reporting and analysis at zero marginal cost. The single biggest time sink in paid media management is compiling data — from Ads Manager, Google Analytics, the attribution tool, the CRM — into something a human can read. MCPs collapse this to a prompt. A weekly performance summary that used to take an analyst two hours now takes 40 seconds.
- Rules-based optimisations, executed reliably. Pause underperforming ad sets when CPA exceeds threshold. Scale budget on ROAS above target. Adjust bids at daypart. These are mechanical tasks that agencies charge coordination overhead to manage. MCPs execute them without coordination friction.
- Speed of iteration on copy and creative briefs. An AI agent with MCP access can read live performance data, identify which ad copy angles are winning, and generate new variants briefed off real signal — not the account manager's intuition. The creative feedback loop compresses from weeks to hours.
- Campaign setup for repeatable playbooks. If you're running the same campaign structure across multiple markets — same targeting logic, same funnel, different localisation — MCPs make that templating trivially automated. The agency equivalent charges for this as setup hours.
The cons: where MCPs expose you to risk.
- Strategy is still a human job. An MCP can execute a bid strategy but it cannot decide what the bid strategy should be. It cannot tell you whether you should be in Google Search at all right now, or whether Meta is structurally the wrong channel for your CAC target and business model. That judgment requires context the MCP doesn't have.
- Creative quality is not a function call. MCPs make creative iteration faster, but the judgment about what makes a piece of creative land — what the brand sounds like, what the audience actually responds to emotionally, what the pixel-level decisions mean — that's craft, not automation. Copy generated by an LLM against performance data alone tends to optimise toward click through and away from brand fit.
- Attribution in complex markets is still broken. MCPs read what the platform reports. Platform-reported attribution has always been optimistic. In markets with fragmented data infrastructure, short cookie lifespans, and mobile-first behaviour, the reported numbers diverge significantly from reality. A senior media buyer knows this and adjusts mentally. An MCP optimises against the reported numbers as truth.
- Write access magnifies mistakes. This is the one that keeps me careful. An MCP with write access to your ad account can pause all your campaigns if you give it the wrong instruction. It can drain a budget in 90 minutes if a scaling rule is misconfigured. The blast radius of an AI agent error in paid media is proportional to spend. That's not an argument against MCPs — it's an argument for tight guardrails and human approval gates on any action with financial consequence.
MCPs don't replace paid media expertise. They replace the coordination overhead and execution labour that expertise has historically been bundled with. If your agency's value is in that overhead, they should be worried. If their value is in judgment, strategy, and market knowledge, they're safe — and more useful than ever with MCPs handling the mechanical layer.
What I'm seeing in African markets specifically.
The Africa context makes this more nuanced, not less.
The agency alternative is often not very good. In most Western markets the "should I fire my agency for MCPs" question assumes you have a capable agency to fire. In many African markets — particularly outside South Africa and Nigeria — the quality of paid media management is thin. The median agency is executing based on global best practices that weren't built for local CPMs, local audience behaviour, or local attribution realities. An AI agent with MCPs and the right guardrails can credibly replace that — not because it's better, but because the baseline is low.
WhatsApp complicates everything. A meaningful chunk of paid media conversions in Africa don't happen in-app or on a landing page — they flow into WhatsApp and disappear from the attribution view. MCPs read what the platform reports. They can't see the WhatsApp conversion. This means any MCP-driven optimisation that's scaling toward reported conversion will, in many African market contexts, be optimising toward a proxy that doesn't accurately represent business outcome. You need a human who understands this and adjusts.
CPMs are structurally different. African CPMs on Meta and Google are dramatically lower than Western equivalents — sometimes by an order of magnitude. This changes the economics of media buying. It changes the creative strategy. It changes the bid strategy. Most of the optimisation logic baked into MCP-executed rules is calibrated for high-CPM, high-competition markets. Running those rules in a low-CPM, lower-competition African context will produce suboptimal outcomes without recalibration.
Regulatory and payment rails add complexity. Running paid media for a fintech in Nigeria means working around NITDA guidelines, platform-level fintech advertising restrictions, and the reality that a significant portion of the conversion funnel runs through USSD. None of that is encoded in an MCP. An agent that doesn't know it will optimise around the wrong constraints.
The verdict, practically.
For a seed-stage African startup with a small budget and no in-house media expertise: MCPs lower the barrier to entry and remove the need for a generalist agency that was charging for overhead you don't need yet. Set them up with tight budget caps, manual approval on any spend-changing action, and weekly human review. You'll get 80% of the tactical execution value at a fraction of the cost.
For a growth-stage business doing meaningful paid media spend in African markets: MCPs should be in your stack for reporting, rules-based optimisation, and creative iteration. They should not be making your strategy calls. The best setup I've seen is MCPs doing the mechanical layer — pulling data, executing tested rules, running copy variants — and a senior media operator doing 4 hours per week of strategy, channel allocation, and anomaly review. That's the new "agency" configuration.
The agencies that survive this transition are the ones who already knew their value wasn't in the dashboard management.