Digiday Publishing Summit: Prices rise Aug. 5

Hear from execs at The New York Times, Thomson Reuters, Trusted Media Brands and many others

SECURE YOUR SEAT

WTF is Model Context Protocol (MCP) and why should publishers care?

As a Digiday+ member, you were able to access this article early through the Digiday+ Story Preview email. See other exclusives or manage your account.This article was provided as an exclusive preview for Digiday+ members, who were able to access it early. Check out the other features included with Digiday+ to help you stay ahead

Talk to any expert in the AI field, and they’ll tell you that the agentic web is coming. It’s a vision of the internet where AI agents make decisions and perform tasks on behalf of users. But that means websites may need to be refitted so that they’re easily understood by AI.

That’s where Model Context Protocol comes in.

What does any of this mean for publishers? In the future, publishers may need to figure out how to make their content accessible to AI agents – and their users – if they want their content surfaced in AI tools that people will use more and more as gateways to the web. Publishers will have to figure out how to rewrite their websites for an agentic web. 

But how soon – and how they do that – remains to be seen. There are a number of technical protocols and frameworks being developed now that can help publishers turn their sites agentic. One of them gaining more attention is called Model Context Protocol.

We’ll dig into what it means, and how it applies to publishers.

WTF is MCP?

Model Context Protocol, or MCP, standardizes how an AI application connects to external sources. It was developed by AI company Anthropic and released in open-source last November. 

It’s kind of like robots.txt for AI, in which publishers can structure how content is shared with AI systems. Developers can put data into a server and make it accessible to AI systems (and, on the flip side, lock away data that’s not in that server from AI agents). Another way of thinking of it is an API for AI.

LLMs are designed for processing and generating natural language, so if you want to create a custom GPT or get your LLM to talk to an application, “there’s not really an easy way to do it without translating it into text,” said Burhan Hamid, co-founder of AI video ad platform streamr.ai and former CTO at Time.

Enter MCP. Data can go into an MCP server, making it easier for LLMs to understand.

“It’s middleware. You can work with an API directly or you can build middleware that lets AI agents work with your APIs,” Hamid said.

What problem does MCP aim to solve?

The main one is integration. AI agents can’t talk to each other easily because they’re built on different infrastructures. MCP is a framework that allows these different AI agents to have a common language. It standardizes the process of AI systems connecting with external data sources.

Ok, but why does this matter to publishers?

We won’t get too in the weeds on the other technical details around MCP, so that we can focus on the important question: does any of this matter to publishers?

In theory, yes. It all goes back to how the agentic web will continue to develop. Right now, AI agents access and crawl websites, but that might not be how AI interacts with sites in the future. MCP servers can give publishers control over how content is shared with AI systems, and what parts of its site (and datasets) are excluded from that dataset.

And because publishers could share content that is licensed to an AI system, but not the content that isn’t, they could prevent full-spectrum access to their content. MCP has the potential to help publishers gain back some leverage in monetizing that content, too.

For example, TollBit, a data marketplace for publishers and AI companies, offers publishers the ability to build MCP servers. AI agents can search for information on a publisher’s site that is available to them through an MCP server. And publishers can charge those AI agents for those queries.

“It gives an additional way for publishers to expose their data to the growing segment of AI apps,” said Toshit Panigrahi, co-founder and CEO of TollBit. 

What are some use cases for publishers?

They’re still being developed. One of them is the site search (and pay per query model) explained above. A publisher could create an MCP server with a dataset (with premium or paywalled content) and monetize that.

Streamr is working on launching media plans on MCP servers – with the idea being that AI agents can go out and “talk” to five different ad servers and come back with all the information needed to build a campaign, Hamid said.

And MCPs could help users access publishers’ content in the AI tools of their choosing, said Nicholas Diakopoulos, computational journalism professor at Northwestern University. 

“[MCP] lets you index all of your content so it can get so it can be a plugin into a chat box. And so imagine a world where I subscribe to The Atlantic and The New York Times… and along with my subscriptions, I get access to the Model Context Protocol server that lets me bring that content into my interface of choice [such as ChatGPT or Gemini],” he said.

Diakopoulos called this “content portability,” in which a user can access content through an AI tool rather than going to a website. “That kind of user-oriented thinking may have some value [for publishers],” he said.

Sounds great. What’s the catch?

In practice, it’s unclear if or how much MCP will be picked up as the standard for the agentic web.

Microsoft created its own protocol called NLWeb in May, and Google launched its Agent2Agent Protocol in April.

“If everyone has their own standard, nobody has a standard,” Hamid said.

He isn’t convinced that MCP servers can provide information to AI agents that they don’t already have access to from crawling sites (which, ultimately, is at the heart of all of this).

“You almost have to build a product for the agents to be able to consume that’s valuable to the agents – more valuable than just scraping your site,” Hamid said. “TBD if that framework is going to be effective and how long it’ll take to build interesting applications with it.”

It’s unclear how many publishers are building on these frameworks. One commercial exec at a large digital publisher said they’re having discussions about this now, and developing an AI agent for other agents for its site.

“The agentic web and the various protocols and how you represent yourself – we’re having some conversations [about that],” they said. “We are very much thinking about what it means in terms of our discoverability and the structure of our data and the usefulness of the data to the agents across the agentic web.”

Until there’s a market for monetizing content through MCP servers, it’s unlikely to be a priority for publishers right now.

“I think it’s an interesting tool that you can experiment with, but… unless you have a mobile app that can use MCP on the back end, just to interact with new services in a way that you wouldn’t be able to do before, I don’t think this is [top of publishers’ mind] with all the other things they have to prioritize right now,” Hamid said.

More in Media

Inside IAB Tech Lab’s meeting with publishers to confront the AI era

Digiday’s Sara Guaglione and Seb Joseph share their reporting on IAB Tech Lab, meeting with more than 80 publishers on AI issues.

Ad Tech Briefing: Lines are being drawn in Amazon and Google’s evolving rivalry 

As advertising nears a double-digit revenue figure for Amazon, a Cold War between it and Google is starting to emerge. 

Creators are ‘doubling’ their rates as they capitalize on brands’ growing interest

Seven influencer marketers tsaid that creators’ rates had risen noticeably across all platforms, although they provided a range of figures regarding the specific amount.