Key Takeaways
- Amazon launched the Ads MCP Server on February 2nd, 2025, giving AI agents direct access to campaign management through natural language commands.
- Agencies using AI workflows report 70-80% reductions in campaign setup and management time, according to research from WMedia.
- A three-layer system of Projects, Skills, and Connectors transforms an agency from a service business into a software-like operation.
- AI-powered agencies can manage double the client load with the same headcount, creating profit margins around 55% versus 16% for manual agencies.
- The compounding advantage means agencies that started building AI workflows 12 months ago cannot be caught up to quickly by agencies starting today.
- Gartner projects that by 2030, most organizations will operate with radically smaller teams amplified by AI, making manual-first agencies structurally uncompetitive.
General Summary
Amazon’s launch of the Ads MCP Server in February 2025 removed the last major technical barrier between AI agents and live campaign management. The Model Context Protocol, an open standard built by Anthropic, lets tools like Claude talk directly to Amazon’s advertising API. Tasks that required 15-20 minutes of manual navigation now execute in a single sentence. Agencies that understand this shift are restructuring how they operate. Those that do not are watching their cost base become incompatible with what the market will bear. Amazon ad revenue hit $68 billion in 2024, up 22% year-over-year in Q4 alone, and average CPC is rising. The margin pressure was already real. AI changes who absorbs it.
Extractive Summary
Amazon’s MCP Server connects AI agents directly to the advertising API, replacing multi-step manual workflows with single natural language commands. Agencies that adopt AI workflows report 70-80% reductions in campaign management time. A three-layer system built from Projects, Skills, and Connectors turns repeatable agency work into automated, consistent outputs. Anthropic’s analysis of 100,000 real-world tasks found AI reduces completion time by roughly 80% on average. The compounding nature of this advantage means the gap between early and late adopters widens every week. Gartner estimates 40% of enterprise applications will have AI agents embedded by the end of 2025.
Abstractive Summary
What Amazon did on February 2nd was not release a feature. It removed a category of friction that kept AI automation out of advertising management. Agencies have spent years differentiating on the quality of their human judgment. That differentiation still matters. But the cost of delivering it is about to diverge sharply between agencies that automate routine work and those that staff it. The structural shift is similar to what accounting software did to bookkeeping or what Shopify did to storefront development. The underlying skill remains valuable. The hours required to apply it collapse. Agencies that recognize this early and build systems around it will not just survive the compression. They will grow into the space left by agencies that do not.
What did Amazon launch on February 2nd, 2025, and why does it matter for agencies?
Amazon launched the Ads MCP Server at the IAB Annual Leadership Meeting, giving AI agents direct access to campaign management through natural language commands. MCP stands for Model Context Protocol, an open standard built by Anthropic that lets AI tools talk to Amazon’s advertising API without custom integrations.
Before this, connecting AI to Amazon Ads required developers, API wrappers, and months of setup. That technical cost kept most agencies and sellers out. The server removes it entirely.
Any Amazon Ads partner with active API credentials can now connect. The server works with Claude, ChatGPT, and Gemini. Access costs roughly 10 GBP a week through tools like Marketplace Ad Pros, plus whatever an agency already pays for its AI platform.
The practical result is striking. Creating a Sponsored Products campaign used to mean logging into Campaign Manager, clicking through menus, setting up ad groups, adding keywords, configuring bids, and setting a budget. That process takes 15-20 minutes when someone already knows the interface. Now it takes one sentence: “Create a Sponsored Products campaign for this ASIN with a $50 daily budget.”
Expanding that campaign to Canada takes four words: “Expand that campaign to Canada.” Pausing every campaign with ROAS under 2 across an entire account used to mean clicking through each one individually. Now it is one command.
The server also connects to Amazon Marketing Cloud. Natural language queries translate into SQL analytics, covering attribution data, audience building, and cross-channel measurement. That capability alone changes how agencies deliver reporting value to clients.
Before the MCP Server, agencies that wanted AI to touch their Amazon accounts faced a familiar sequence: find a developer, budget three to six months of build time, manage API versioning, and maintain the integration every time Amazon pushed an update. The setup cost was high enough that most agencies never started. That barrier is gone. What replaced it is open access, minimal cost, and tools that work with AI platforms agencies are likely already using.
The agencies paying attention to this are not waiting for a mature product to arrive. They are building now, accepting rough edges, and accumulating the institutional knowledge that makes their systems proprietary. The agencies waiting for the tool to be perfect are watching the head start grow.
Why can manual agencies no longer compete on price, speed, or quality?
Manual agencies face a structural cost problem that AI adoption has made permanent. Amazon’s ad revenue reached $68 billion in 2024, a 22% increase year-over-year in Q4. Average CPC hit $1.12 in 2025, up 15% from the previous year, and is projected to reach $1.25. More advertisers are competing for the same placements.
The economics become clear when two equally sized agencies are compared side by side. Both have five analysts. Both manage 15 clients. One adopts AI workflows. The other keeps doing things manually.
Six months later, the AI-powered agency manages 30 clients with the same five people. The manual agency is still at 15 and needs to hire five to eight more people to grow. Profit margins reflect this directly: roughly 55% for the AI agency, around 16% for the manual one.
Same talent. Same market. Entirely different unit economics.
The AI agency can now undercut manual agency pricing by 20% and remain more profitable. Or it can match the pricing and deliver twice the strategic output: more competitive analysis, more frequent reporting, more proactive optimization. The manual agency faces a binary choice. Match the lower pricing and bleed margin. Keep the higher pricing and watch clients leave.
Research from WMedia found 70-80% time reductions in campaign setup and management after agencies adopted AI agents. That efficiency gap is not coming. It is already operating in the market.
The CPC increase compounds the problem. Every year CPCs rise, manual agencies need more analyst hours to maintain the same return on ad spend for clients. They either absorb the cost, raise fees, or accept declining performance. AI-powered agencies face no such constraint. Their cost per managed account falls as their systems mature, even as CPCs rise. The divergence in client outcomes widens alongside the divergence in operating costs.
There is also a quality ceiling in manual operations that does not exist in AI-augmented ones. A manual analyst can review one campaign structure at a time. An AI-assisted analyst can review all of them simultaneously, cross-reference against historical performance, check competitor activity, and surface the three most impactful actions. The output is not slightly better. It is categorically different in depth and speed.
What is the three-layer system that turns an agency into a software operation?
The three-layer system consists of Projects, Skills, and Connectors, each compounding the others to replace manual throughput with automated, consistent output. Understanding each layer separately clarifies how they combine.
What do Projects do inside the system?
Projects give Claude persistent client context across every conversation. Each client gets a dedicated Claude Project that holds their goals, history, constraints, brand voice, and competitive landscape. Every conversation starts with full context rather than starting from zero.
Without Projects, every new session requires re-briefing. An analyst spends the first ten minutes of every conversation re-establishing what the client cares about, what was tried last month, and what the account constraints are. Projects eliminate that cost entirely.
What are Skills and how do they standardize quality across a team?
Skills are automated workflows that encode a senior analyst’s methodology into a repeatable process any team member can run. A senior person builds a Weekly Campaign Review skill once. Every analyst then runs that skill and gets the same quality output, regardless of their experience level.
In a manual agency, quality varies by seniority. A senior analyst identifies three optimization opportunities. A junior analyst finds one and misses two. Same client, different outcomes. Skills close that gap. Someone who started the previous week delivers senior-level analysis in their first week because the thinking is already encoded in the skill.
Skills also make scaling cheaper. Adding a new team member no longer requires six months before they reach full productivity. They read the client Project, run the existing Skills, and deliver immediately.
What do Connectors plug into the system?
Connectors link Claude directly to external tools: Amazon Ads data through Marketplace Ad Pros, competitor research through Ahrefs, meeting transcripts through Fireflies, client communication through Slack. Each Connector eliminates a manual data-gathering step.
The combined effect is visible in a concrete example. A client asks how they compare to competitors. A manual agency exports Amazon data, opens Ahrefs separately, researches competitors manually, compiles a spreadsheet, writes the analysis, builds a presentation, and emails it. That takes two to three hours. So it happens monthly, at best. With Connectors in place, the same task takes five minutes. It happens weekly. Clients notice the difference.
This three-layer system is proprietary. Each skill built, each project configured, each connector wired represents institutional knowledge a competitor cannot purchase. They would need months of iteration to replicate it, starting from the same 2x baseline every early adopter began with.
What does the data show about real productivity gains from AI in advertising?
Anthropic analyzed 100,000 real-world Claude conversations and found AI reduces task completion time by roughly 80% on average. Tasks that took a professional 1.4 hours dropped to a fraction of that. These were not lab conditions. They were actual tasks people completed in daily work.
Anthropic’s own internal team uses Claude in 60% of their daily work and reports a 50% productivity increase. The most significant finding: 27% of the work completed with Claude consisted of tasks that would not have been done at all otherwise. Projects that were not worth the time manually became viable with AI assistance.
Independent reports confirm the pattern. On LinkedIn, Shuqing Ke documented that after connecting to the Amazon Ads MCP Server, client reporting time dropped from 5 hours to 30 minutes. Hasaam Bhatti built a multi-agent system for Amazon PPC using 4 analyst agents, 3 reporter agents, and 1 orchestrator. His morning routine went from 4 hours of report-pulling and bid adjustments to 15 minutes.
On Reddit, an ad manager described automating paid ads reporting across Google, Meta, and LinkedIn. Mondays that used to take three to four hours of exporting and copy-pasting dropped to about 30 minutes. The same pattern applies directly to Amazon Ads reporting.
Matt Anderson, writing on LinkedIn, framed the existential risk plainly: 75% of what an Amazon advertising agency sells could become obsolete. His warning was that clients will eventually realize a 50 GBP monthly license lets AI handle most of this in real time. When they do, agencies that have not already restructured around value creation will have no answer.
These are not projections about a distant future. The people documenting these results are working Amazon Ads practitioners describing their current routine. The time savings are not theoretical. They are already inside agencies that made the decision to connect and build.
For agencies that have not yet started, the most useful framing is not what AI could do but what agencies using AI are doing right now that their manual counterparts are completing in three times the hours. The answer covers most of the work.
How has Clear Ads built an AI-powered agency system internally?
Clear Ads went from producing 11 pieces of content per month to 418 with the same working hours and the same team. That is a 38x output multiplier. The increase did not come at the cost of quality. Every piece runs through Skills that encode the agency’s best thinking, so quality increased alongside volume.
The operational shift is visible in how analyst time is allocated. Before AI workflows, analysts spent roughly 70% of their time on routine work: pulling data, running basic optimizations, building reports. Work that has to happen but does not require strategic judgment. Now that split has reversed. Approximately 20% of time goes to routine analysis, because Claude handles most of it. About 50% goes to strategic work. The remaining 30% goes to client relationships.
Compared to a manual agency running the same number of accounts, this means more than three times as much time spent on strategy and five times more time on relationships. The output clients receive has more depth. They also hear from their account managers more often.
Scaling from 20 clients to 40 now requires two to three new hires, mostly for account management. The Skills are already built. The quality is already automated. A new hire reads the client Projects, runs the existing Skills, and delivers senior-level analysis in their first week.
The compounding effect is significant. In month one, the system ran roughly 2x faster than manual processes. By month six, it reached 5x. Current performance sits at 8-10x on most workflows. A competitor starting today begins at 2x. They cannot skip the iteration required to close that gap. Every week the system runs, it refines further.
Why does starting late cost more than it appears?
The compounding nature of AI adoption means early movers do not just work faster: they learn faster. Every skill built, every prompt refined, every workflow tested is institutional knowledge that accumulates. The gap between an agency that started 12 months ago and one starting today is not 12 months of lag time. It is 12 months of compounding iteration.
Gartner estimates 40% of enterprise applications will have AI agents embedded by the end of 2025. Not by 2030. This year. Spencer Stuart surveyed senior marketers and found 36% expect to reduce headcount in the next 12-24 months because of AI. Among companies doing $20 billion or more in revenue, the likelihood of having already cut roles is two and a half times higher than smaller companies.
By 2030, Gartner projects most organizations will operate with radically smaller teams amplified by AI. IDC found that 70% of CEOs are already pursuing revenue growth without adding headcount. For agencies, the implication is direct. Compete as a lean, AI-amplified team or compete against those teams while carrying three times the overhead.
The window to build a compounding advantage still exists. It is narrowing. Agencies that treat this as a future consideration are already behind agencies that treated it as an immediate operational priority 12 months ago. The question is not whether to start. It is how far behind the starting line is.
What should an agency do first to begin building an AI-powered operation?
The first step is connecting to the Amazon Ads MCP Server using existing API credentials, which gives immediate access to natural language campaign management. The total cost for data access and MCP hosting runs around 10 GBP per week through tools like Marketplace Ad Pros, plus an existing AI platform subscription.
From there, the priority is building the first Project. Pick one client. Document their goals, account history, brand constraints, and competitive context in a Claude Project. Run a single task, such as a weekly campaign review, through that Project and compare the output to what a manual review produces.
The first Skill worth building is whichever task consumes the most analyst hours per week. For most agencies, this is reporting. Map out every step of the current reporting process. Encode it as a Skill. Run it. Refine the output. Once one Skill works reliably, the second is faster to build because the pattern is established.
Connectors come after Projects and Skills are functioning. Add Marketplace Ad Pros for live Amazon data, then Ahrefs for competitive research. Each Connector reduces a manual data-gathering step and feeds richer context into every Skill.
The key distinction is that this system is not purchased. It is built. Each agency’s version reflects its own methodology, its own client base, its own competitive positioning. That is what makes it proprietary. Two agencies can use identical tools and produce entirely different systems based on what they encode into their Projects and Skills. Start building now. The compounding starts on day one.
There is a common objection at this point: the agency is already busy running accounts and does not have time to build new infrastructure. This framing inverts the problem. The agencies that built these systems did so while managing existing clients. They started with one skill on one client. The first build took a few hours. The second took less. By the time the system covered five workflows, the time saved outpaced the time invested in building.
The alternative is continuing to compete on the same terms as before while the competitive landscape restructures around AI-powered operations. Manual agencies are not standing still while AI agencies move ahead. They are actively losing ground, because the gap compounds in both directions: AI agencies get faster while manual agencies stay the same.
Amazon’s advertising platform will not stop evolving. New placements, new targeting options, new reporting surfaces, and new algorithm updates arrive continuously. In a manual operation, each new capability requires analyst hours to learn, test, and implement across accounts. In an AI-augmented operation, a new Skill captures the methodology once and every account benefits immediately. The ability to respond to platform changes at scale is, by itself, a durable competitive advantage.
Amazon’s MCP Server connects Claude directly to ad campaigns. Agencies using AI now manage 2x clients at 55% margins. Here’s what manual agencies face.
