In this deep-dive, Dmitry Sverdlik, CEO at AI and software development company Xenoss takes a closer look at how AI agents, and the protocols that power them, are starting to reshape the web as we know it. Behind the scenes, a quiet revolution is underway, and it’s being led by something called the Model Context Protocol (MCP). If you’re in ad tech or working with AI, this is one shift you’ll want to pay attention to.
When OpenAI released Operator, it signalled a seismic shift in how we interact with the web; people quickly started unraveling the implications of AI agents.
Aside from letting tools like Operator streamline work, the web where agents do the clicking and searching starts to change. If no one’s searching or browsing, there’s no one to monetise. The web as we know it, built on pageviews and paid clicks, could be on the verge of collapse.
At the same time, the Federal Court was busy with Google’s anti-monopoly lawsuit, which has the potential to unravel the Internet further. The restrictions the DOJ is seeking to place on the company will hit its multibillion-dollar partnerships with device manufacturers for keeping Google the default search engine, break up Google’s ad tech ecosystems, and encourage competition in the search space.
If usage patterns change and Google’s stake in the open web declines, its incentive to maintain it could vanish. And if both demand and support fall away, the version of the internet we’ve relied on for 20 years may not survive this decade.
But if one door closes, another will open. Behind that door, I expect to see the web of AI agents.
Enter the Agentic Web
At Microsoft’s Build 2025 Developer Conference, the company pitched a proposal for “The Open Agentic Web”. The new web rests on two foundations: smarter agents that are easier to build and deploy, and a standardised connection layer that connects them.
To support this, Microsoft is releasing new developer-facing infrastructure. Tools like AI Foundry Agent Service and the AI Copilot aim to accelerate agent development. But the real infrastructure breakthrough is the standardised layer powering their communication: the Model Context Protocol, or MCP.
Understanding MCP: The Protocol Powering the Agentic Web
When MCP was released by Anthropic in late 2024, it somehow flew under the radar of the AI community. By early 2025, that changed fast; the protocol had a community of early adopters, but its popularity took off when Sam Altman announced OpenAI enabled MCP support.
Other big names followed: Google DeepMind, Microsoft, and thousands of enterprise organisations now support the protocol.
Today, MCP adoption is growing so rapidly that it’s on track to outpace GPT usage across enterprise environments.
That’s why I believe any team using AI across its advertising value chain should care about MCP and try to make the most out of it.
To see what’s at stake, consider how agents work without it.
Every useful agent needs data and context. That’s why developers connect models to data sources: CRM records, DSP/SSP logs, historical campaign performance, product catalogs, and inventory tables.
More available data sources lead to not just smarter, but context-sensitive agents that understand your team’s needs, goals, and limitations. It’s this awareness that can help AI agents make a jump from simply “intelligent” to “helpful”.
But adding a data source is not a one-click task. Each source needs a unique API, an integration, and custom logic. Setting them up would take time and developer work, but teams eager to tap into machine learning had to accept these constraints because there was no other way.
With MCP, There is Another Way
MCP finally offers a unified way to connect AI agents to data. It can request data from any tool, get instant results, and build complex action chains without defining where the data comes from beforehand.
I think the AI community is spot-on in comparing MCP to the USB-C adapter for data sources. Both are standardised connectors that can be plugged into any source.
Here’s what that looks like in practice.
An AI tool, or MCP host, uses the protocol to connect to an MCP server. MCP servers enable access to data sources such as internal databases, CRM records, AdTech platform logs, and others.
They also run commands, access the company’s prompt library to find the right prompt for communicating with other AI tools, handle errors, and send feedback to the AI agent.
Importantly, MCP isn’t limited to a particular system or location. It can pull from both local and remote sources, giving AI agents full visibility across the stack.
Why MCP is Winning
Most ad tech leaders know the upside to building with AI and even have use cases lined up: creative optimisation, intelligent bid engines, granular targeting, and many more.
Yet, they are uncertain about committing to any of them, mostly due to three reasons: slow adoption, lack of interconnectivity within the ecosystem, and unpredictable quality of AI performance.
LLMs and generative AI have made machine learning significantly more accessible and have already helped solve these problems. But I am excited about MCP because it takes development speed, connectivity, and model performance to a new level.
MCP redefines the pace and potential of what’s possible. Here’s how.
More Efficient Processes and Faster Project Release
Data is the foundation of any successful AI initiative. The more sources you have, the more confident you can be that your AI delivers what you expect.
Pre-MCP, adding a new data source to machine learning would require engineering expertise and domain knowledge. Now it’s enough to connect an AI tool to an MCP server (which you don’t even have to build due to dozens of emerging marketplaces and open-source libraries).
With data-related friction out of the picture, piloting a new AI use case will take weeks, which would have taken three to four months. This is a big deal in today’s market, where the winning teams are ready to seize opportunities immediately.
Improved Collaboration Within the Ecosystem
In the ad tech industry, we talk a lot about data collaboration, but this spirited vision usually fails to pass the reality check of technical implementation.
With MCP, interconnectedness between the sell-side, buy-side, and vendors is no longer a pipe dream. The protocol makes sharing data and helping each other build better AI use cases feasible and realistic.
Smarter AI Agents
One of MCP’s most important promises, and the one I don’t see getting the limelight it deserves, is creating a “collective consciousness” among AI agents. We’ve already seen LLMs develop the idea of internal memory that helps the model track user preferences and retrieve data from multiple chats.
Projects like OpenMemory MCP take this further, creating a shared memory for different LLMs and AI agents. Now, your Performance Management agent will remember and retrieve the output of your Campaign Management agent, allowing everyone on your AI team to be on the same page.
Shared memory is game-changing because it transforms one-off AI tools teams use to automate individual tasks into fully integrated teams that can run complex workflows without human supervision. Hence, I expect Anthropic’s MCP and Google’s A2A to be the key elements empowering the multi-agent systems I discussed in my previous write-up.
Ad Ops Before and After MCP
While the benefits of MCP apply across industries, I believe its impact on ad ops will be especially significant. From improving inter-system connectivity to enabling smarter, self-directed agents, the protocol addresses long-standing friction points in campaign execution and optimisation.
That’s why it’s worth zooming in on what MCP unlocks specifically for ad ops teams in day-to-day workflows.
To that end, I outlined five use cases where the speed and performance gains of MCP really come through.
Cross-Platform Creative Optimisation
Before AI agents and MCP, creative teams had to deal with cross-platform fragmentation. They had to manually switch between tools: Performance Max for Google, Advantage+ for Meta, custom rules in DSP, and many more. Each platform was siloed, so teams had to build workarounds like unified dashboards that stitched everything together in a single view.
With agents connected via MCP, ad ops teams will be able to create a central intelligence hub. Agents from Meta, Google, Amazon, TikTok, and other platforms will communicate via the protocol and share creative performance, placement effectiveness, and other campaign data.
In this workflow, humans no longer oversee the execution but focus on big-picture strategic goals.
Media Planning is Unified
Media planning is still supported by tons of manual data gathering and entry. Teams have to pull up reports from disparate sources (CRMs, DSPs, Google Ads, Meta Business Suite, etc.).
Besides, it’s not common for media planners to have robust analytics toolsets to aggregate this data and spot helicopter-view trends that optimise media plans.
MCP-powered AI agents would enable and automate this big-picture view. The protocol would allow AI agents to automatically go through CRM, campaign, historical performance, and inventory data. Then, an AI service would automatically build a media plan based on real-time demographics analysis, competitor landscape, inventory availability, and campaign budget.
Alternative to Third-party Cookies
Considering the impact of cookies on the economics of advertising and a string of failed attempts by Google to replace them, it’s unlikely that everyone will switch to an AI-driven standardised alternative in the next few years.
But theoretically, AI agents communicating via MCP can lay the foundation for privacy-preserving and data-rich tracking.
Instead of tracking a user’s on-page behaviour, interconnected agents will use contextual memory retrieval to recall user-consented past interactions (e.g., types of content consumed). This will unfold across the entire web in real time, with less reliance on third-party IDs and better contextual relevance.
AI Search and Retailer Feed Convergence
In the good old Internet, making a purchase happens across multiple touchpoints: search engines, social media, and, finally, a retailer’s own platform, where the checkout is complete.
These context changes increase drop-offs and have users switching between retailers depending on what they are looking for.
As soon as 2-3 years from now, I expect this model to be largely substituted by agent-first purchasing journey. Soon, a user will only need to make a request via ChatGPT, intelligent hardware, or other AI interface, and agents will take it from there.
MCP will allow them to pull up any data source to explore options, check reviews, and even complete the payment by accessing a shopper’s payment apps.
Retailer-owned media will likely lose ground to AI marketplaces, but the bottom line will be a win-win. Shoppers will be enjoying an improved user experience, and retailers will get frictionless customer acquisition.
Direct Buys Go Programmatic
Programmatic helped automate most inventory, but direct deals still verge on humans exchanging email threads and creating rate card spreadsheets to reach a deal.
In an agent-to-agent negotiation, a buyer’s AI agent will use MCP access to audience, budget, and brand safety data. In contrast, a publisher’s agent will keep track of premium inventory availability and pricing trends.
A shift from highly manual to automated negotiations will cut time to deal, improve inventory availability, and help publishers tap into dynamic pricing now that they have access to the latest market data at all times.
These applications are only the tip of the iceberg of the revolution AI agents supported by MCP can bring about in ad ops. I believe they can help streamline, if not take over, nearly every workflow in media buying, campaign management, and performance tracking.
But disruptive change can be unsettling to team leaders.
In such pivotal moments, their strategic insight and skill in building their AI adoption roadmap will brand their teams as leaders or laggards for years to come.
That’s why having a roadmap for integrating up-and-coming tech like MCP should be top-of-the-mind for AdOps leaders even if they don’t directly interact with the engineering side of the business.
Here are the considerations I would recommend ad ops teams to keep in mind if they want to use MCP-supported AI agents.
Switch From Monolithic Tools to Modular Workflows
MCP breaks complex workflows into smaller AI services, each of which connects to a dedicated MCP server.
This modular and interconnected nature has much in common with microservice architecture, a software design principle that breaks software into smaller, independent moving parts.
Introducing AI features into your ad tech stack as autonomous, loosely connected components will make it easier to decouple each service and connect dedicated MCP servers.
Set Up Authentication and Define Access Control
Because MCP is still early in its lifecycle, it doesn’t yet offer built-in authentication standards. That puts the responsibility on adopters.
As a leader, make sure your IT team sets up the necessary guardrails—secure authentication protocols (like OAuth), tenant-aware access, and permission logic tailored to how your agents will interact with sensitive data.
Get this right early, and you’ll save significant time and risk later.
Build AI Adoption with MCP Accessibility in Mind
Whether AdOps teams are using off-the-shelf LLMs or building in-house, MCP support is a must. While the protocol is only starting to gain traction, it still seems like the most promising candidate for the title of “AI’s universal language.”
Replacing legacy systems with MCP-enabled, intelligent platforms modernises your stack and puts you in sync with the direction the industry is already heading.
The Bottom Line
Together with intelligent agents and inter-agent protocols like A2A, MCP is fuelling a fundamental shift in how ad tech operates. Within just a few years, ad ops teams will no longer be tool operators; they’ll be strategic directors of autonomous, data-driven AI units.
As execution becomes automated, the human role will evolve. Teams will spend less time managing dashboards and more time shaping long-term vision, developing creative strategies, and deepening their understanding of the audience.
I don’t know whether Web 2.0, as we know it, is truly “dying.” But I do believe this: if we let AI agents handle the spreadsheets and the clicking while humans focus on creativity and strategic direction, the next version of the Internet will deliver more inspired, relevant, and respectful advertising.
And if that’s where we’re headed, I think it’s a future worth building.
