TL;DR: The Model Context Protocol (MCP), Anthropic’s open standard for connecting AI agents to external tools, is crossing over from developer circles into marketing departments. Amazon Ads launched an MCP server in open beta. Early adopters report 67% faster campaign launches. Gartner projects 33% of enterprise software will include agentic AI by 2028. The protocol is positioning itself as the plumbing that connects AI to the $58 billion marketing AI market. Here’s what’s actually happening, who’s building on it, and what it means for your stack.

How the Model Context Protocol Is Reshaping Marketing Workflows

Marketing technology has an integration problem. It’s been around for twenty years and nobody has fixed it properly. The average enterprise marketing team runs somewhere between 20 and 90 tools. Each one has its own API. Each API has its own authentication flow, data format, rate limits, and quirks. Connecting them requires custom middleware, Zapier chains held together with duct tape, or begging the engineering team to build yet another integration instead of actual product features.

Now drop AI agents into that mess.

An AI assistant that writes ad copy? Useful, sure. But an AI assistant that writes the copy, checks it against brand guidelines in your DAM, pulls last quarter’s performance data from the analytics platform, builds the campaign in your ad manager, and pings your team on Slack when it’s live? That’s a different animal. The problem is that building it has required custom engineering for every model-tool combination. Bespoke work, every single time.

The Model Context Protocol is rewriting those economics. And in early 2026, marketers finally started noticing.

A quick primer for the uninitiated

The USB-C analogy works well here. Before USB-C, you needed a different cable for every device. Before MCP, developers needed a different custom integration for every tool they wanted their AI agent to talk to.

Anthropic introduced MCP in November 2024 as an open-source standard defining how AI systems communicate with external tools and data sources. It follows a client-server architecture. Your AI application (the MCP client) connects to MCP servers, which are essentially wrappers around tools like Google Drive, Salesforce, or an ad platform. The server translates requests into API calls the tool understands, then sends structured responses back.

Three components make the protocol tick:

• Tools (model-controlled): Functions the AI can call. “Create campaign.” “Pull report.” That sort of thing.
• Resources (application-controlled): Data the AI can read. Customer segments, performance metrics, creative assets.
• Prompts (user-controlled): Pre-built templates that steer the AI toward using tools and resources effectively.

So what separates this from just building another API wrapper? Scale. Write an MCP server once for a tool, and every MCP-compatible AI agent can use it. Claude, ChatGPT, Gemini, open-source models, whatever comes next. That flips the old N×M integration problem on its head. Instead of building a connector for every model-tool pairing, you build one server per tool and you’re done.

Amazon Ads goes all in

The moment MCP jumped from “interesting developer tool” to “thing marketers need to know about” landed on February 2, 2026. That’s when Amazon Ads launched its MCP server in open beta.

Paula Despins, VP of Ads Measurement at Amazon Ads, framed it around a blunt practical reality: Advertising workflows involve way too many manual steps scattered across too many screens. Campaign creation alone means navigating account settings, building ad groups, configuring targeting, uploading creatives, setting budgets. Each step lives in a different interface or API endpoint. It’s death by a thousand clicks.

The Amazon Ads MCP server bundles all of that into orchestrated workflows an AI agent can fire from a single prompt. Tell Claude to “launch a Sponsored Products campaign for my new running shoes targeting fitness enthusiasts with a $50 daily budget,” and the MCP server handles the multi-step API choreography behind the scenes.

Alex Brockhoff, Senior Technical Product Manager at Amazon Ads, put it simply: MCP adds “a contextual layer that makes those same capabilities usable by AI agents.” Not new functionality. Existing functionality made accessible in a fundamentally different way.

The specifics matter. The server supports:

• Campaign creation, updates, and deletion across Sponsored Products
• Performance reporting and analytics queries
• Account-level settings management
• Billing and financial data access
• Geographic expansion across multiple Amazon marketplaces
• Amazon Marketing Cloud (AMC) query execution through connected LLMs

That last point deserves a closer look. AMC is Amazon’s clean-room analytics environment where advertisers run custom queries against first-party shopping signals. Using it previously required SQL skills and a completely separate workflow from campaign management. With MCP in the picture, an advertiser can ask their AI agent to “show me the overlap between my display audience and my Sponsored Products converters last month” and get an answer without writing SQL or leaving the chat window. That collapses the workflow entirely.

The numbers that got people’s attention

Beta testers haven’t been quiet about their results. According to Stormy AI and Seller Labs, early adopters are reporting:

• 67% faster campaign launches compared to doing it manually
• 300% ROI during beta testing
• 5.6 hours saved per week on reporting and analysis alone, which works out to about 30 working days a year

Those time savings stack up quickly. Think about a mid-size e-commerce brand running campaigns across five Amazon marketplaces. Their advertising manager probably spends half the week pulling reports, comparing numbers across regions, tweaking bids. Offloading the reporting and analysis loop to an MCP-powered agent frees that person up for strategy, creative testing, and market expansion. The stuff that actually moves the revenue needle.

Hector Ai, an AI-powered commerce media platform and Amazon Ads tech partner, has been building at this intersection. Meher Patel, the company’s founder, described their setup as a “hybrid” model where Hector Ai’s proprietary intelligence handles the strategic layer (deciding what campaigns to run, how to optimise them), while Amazon’s MCP server takes care of execution (actually creating and managing the campaigns). MCP makes that kind of two-system orchestration possible without building a brittle, one-off integration.

Beyond Amazon: Where else MCP is showing up

Amazon’s launch grabbed headlines, but MCP’s impact on marketing reaches further than one ad platform. The protocol is model-agnostic and tool-agnostic by design, which means it’s popping up across the martech space in ways worth tracking.

Creative testing at scale. AdSkate built an MCP integration that lets advertisers test campaign creatives against over 1,000 synthetic audiences using conversational AI. Rather than running A/B tests one after another and waiting days for statistical significance, marketers simulate audience response patterns in an AI-powered feedback loop. It won’t replace live testing. But it chops weeks off the iteration cycle before a creative ever reaches real eyeballs.

Cross-platform campaign orchestration. Claude and other MCP-compatible agents can connect to multiple MCP servers at once. In practice, that means a single conversation can touch your entire stack. Pull audience insights from the CDP. Check creative assets in the DAM. Build the campaign in the ad platform. Fire off a Slack message to the team. Reconcile the budget in QuickBooks. Each step hits a different MCP server, but the whole thing feels like one continuous conversation. No tab-switching. No copy-paste between platforms.

Email marketing automation. Teams that wired up MCP for email campaigns have reported 90% fewer campaign errors, per CMSWire. Automated pre-send checks handle rendering across devices, link validation, and personalisation token verification. The kind of grunt work that manual QA misses all the time, especially when people are shipping under deadline pressure on a Friday afternoon.

Customer segmentation in plain English. Enterprise retailers are plugging MCP into their customer data platforms and building segments through conversation. “Show me customers who bought running shoes in Q4, haven’t purchased in 90 days, and have a lifetime value above $500.” That’s a prompt now, not a SQL query or a 12-click odyssey through a segmentation UI.

The protocol space: MCP isn’t alone

It would be misleading to talk about MCP without acknowledging the broader protocol picture. It’s not the only standard competing for a seat at the agentic AI infrastructure table.

Google shipped the Agent-to-Agent (A2A) protocol in April 2025, backed by over 50 technology partners including Salesforce, SAP, and ServiceNow. They donated A2A to the Linux Foundation two months later, which was a clear signal about governance intentions.

The difference between MCP and A2A is structural. MCP handles vertical connections: An AI agent talking to its tools. A2A handles horizontal connections: AI agents talking to each other. They aren’t fighting for the same territory. They occupy different layers of the stack.

Walk through a concrete example. A marketing agent uses MCP to pull campaign performance from Amazon Ads and creative engagement data from your social channels. It spots an underperforming segment. Via A2A, it checks in with a finance agent that confirms budget is available, and a creative agent that generates new ad variations. The marketing agent then uses MCP again to push the refreshed campaign live.

This isn’t hypothetical architecture. It’s the direction that companies like Block (formerly Square), one of the co-founders of the Agentic AI Foundation alongside Anthropic and OpenAI, are actively building toward.

The practical takeaway for marketers: MCP is the protocol you’ll run into first, because it’s the one connecting AI tools to your existing platforms. A2A matters later, once multi-agent orchestration becomes a real part of marketing operations rather than a conference demo.

The enterprise adoption reality check

This is where the excitement needs a cold shower. Sixty percent of AI leaders say integrating agentic AI with legacy systems is their top challenge, per Deloitte. And that’s not a protocol problem. That’s a plumbing problem. Legacy marketing tools weren’t built with MCP in mind. Plenty of them lack modern APIs entirely. Some still run batch processing architectures straight out of 2008.

There’s a wide gap between “MCP exists as a standard” and “my marketing stack actually supports it.” Building an MCP server for a modern SaaS tool with clean REST docs is a weekend project. Building one for an on-premise marketing automation platform from 2015 that runs SOAP and has authentication patterns that make grown developers cry? Different story.

Then there’s trust. Letting an AI agent create campaigns, adjust budgets, and trigger email sends means handing it real authority over real spend. Marketing teams that have spent years building approval workflows and change management processes aren’t going to disable all of that because someone demoed a cool chatbot integration. Nor should they. The protocol itself doesn’t come with guardrails. That responsibility lands squarely on whoever implements it.

As the Digiday piece put it, AI systems in advertising need “stronger embedded context, clearer resource relationships, and domain-aware guardrails for reliable decision-making.” In plain language: The pipe works fine. The brain on the other end of the pipe still needs training wheels before you give it the keys to the media budget.

Follow the money

The push toward MCP-enabled marketing isn’t an isolated trend. It’s riding a wave of capital and momentum that’s hard to dismiss.

The agentic AI market is on track to grow from $7.06 billion in 2025 to $93.2 billion by 2032, a CAGR of 44.6%, per Fortune Business Insights. The AI marketing segment specifically has grown from $6.46 billion in 2018 to $57.99 billion in 2026. That’s 2.5x faster growth than the broader martech industry.

Gartner’s latest projections put it starkly: 33% of enterprise software will include agentic AI by 2028, up from less than 1% in 2024. They also expect 60% of brands to use agentic AI for one-to-one customer interactions by the same year.

All of those projections assume infrastructure exists. You can’t wire agentic AI into enterprise marketing without standardised protocols for agents to access marketing tools. MCP, or something that looks a lot like it, is that infrastructure. Anthropic made MCP open-source for exactly this reason. The protocol gets more valuable as more tools adopt it. Network effects 101.

What this means for your team, right now

Enough theory. Here’s what to actually do with this information.

If you’re an Amazon advertiser, the MCP server is live in open beta. Works with Claude, ChatGPT, and Gemini. You need active Amazon Ads API credentials and an MCP-compatible AI platform. Start with reporting automation. Let the agent prove itself on read-only tasks before you hand it campaign creation privileges. Crawl, walk, run.

If you’re evaluating martech vendors, put “do you have an MCP server?” on your RFP. It’s becoming a real differentiator. A tool with an MCP server means your AI agent can talk to it natively. A tool without one means you’re stuck with custom integrations or doing things by hand. In 2026, that’s a competitive gap.

If you’re building marketing automations, MCP changes the architecture underneath. Instead of wiring point-to-point connections between tools (Tool A fires a webhook to Tool B), you build MCP servers around each tool and let an AI agent orchestrate the whole workflow. The agent becomes the integration layer. That’s not just a technical shift. It changes how you think about workflow design.

If you lead a marketing team, start thinking about what your people need to learn. The teams that figure out how to work with AI agents effectively (good prompting, solid output auditing, smart workflow design) are going to pull ahead of teams that don’t. MCP won’t eliminate anyone’s job. But it will reshape what that job looks like day to day, pushing effort away from clicking through platform UIs and toward strategy and creative thinking.

Where this is headed

2025 was the year MCP gained traction with developers. 2026 is the year it’s crossing into business functions. Marketing, with its sprawling tool ecosystems and constant appetite for automation, was always going to be one of the first to adopt.

But let’s keep perspective. Most marketing MCP implementations right now cluster around a few platforms, with Amazon Ads being the flagship. The dream scenario (a fully MCP-connected stack where an AI agent orchestrates campaigns smoothly across Google Ads, Meta, TikTok, your CDP, CRM, creative suite, and attribution platform) is still aspirational for the vast majority of teams.

The direction is clear, though. You’ve got a protocol that turns any tool into something an AI agent can operate through natural language. You’ve got an industry that spends billions every year on tools that demand manual operation. Those two things are going to collide. The question isn’t whether. It’s how fast.

The marketers who start building familiarity with MCP-powered workflows today, even in small, low-risk experiments, will have a head start when the ecosystem fills out. If the early Amazon Ads data is any indication, that might happen sooner than the conservative forecasts expect.

The standard is open. The servers are shipping. Your stack is the bottleneck now.