AI agents are spreading fast. They perform tasks by navigating the web — but the web was built for humans, not machines.
Millions of hours have gone into crafting user-friendly websites for human eyes (thanks UX experts). But AI agents don’t see. They rely on large language models (LLMs), which work best with simple text — not visual buttons, menus, or layouts encoded with HTML, CSS and JS.
MCP and A2A: how could AI protocols open new revenue streams for media organizations
AI agents are coming, and they want to talk — to each other, to your content, maybe even to your newsroom. Two new protocols — MCP and A2A — could turn this into a business opportunity for publishers.
Let’s unpack what they are, why they matter, and how media companies could turn them into real revenue, surfing the wave of the hype, hoping for long-term returns.
What’s MCP?
Think of MCP (Model Context Protocol) as a universal power plug for AI.
Before, AI tools like ChatGPT or Claude had to be custom-wired to access external data — manual work on APIs and integrations. MCP changes that. It lets any AI model securely plug into tools, databases, or content libraries in a standard way.
By the way, you might be wondering: how is MCP different from a traditional API?
One word — discoverability.
With a classic API, the caller has to know exactly how it works. There’s no way to understand it on the fly. A developer has to manually integrate that API into the LLM’s system ahead of using it.
With MCP, it’s different. Each tool can describe itself directly to the calling LLM — what it does, what inputs it needs, how to use it. No prior setup required. The LLM can explore all available tools, read their MCP contracts, and use them autonomously.
Imagine that the AI assistant is a chef. Without MCP, it’s stuck in an empty kitchen. With MCP, it gets access to your whole fridge — your archives, stats, past interviews, even paywalled reports — and knows how to use it all.
For media companies, that’s a big shift. Suddenly, your editorial archive or photo database can be “opened” to AI agents — not scraped, but properly accessed under your terms. You become part of the AI’s workflow.
Then comes A2A
On April 10th, Google and a group of tech partners launched A2A (Agent-to-Agent Protocol) — a way for AI agents to discover and talk to each other directly. It complements MCP protocol, which gives LLM tools, while A2A gives agents a way to talk to each other.
It’s like email for AI tools. One agent might specialize in legal research, another in news analysis, another in generating reports. With A2A, they can team up — exchanging tasks, updates, even files or charts — without a human in the loop.
How is A2A different from APIs?
APIs follow a deterministic call flow.
You call one endpoint to get all products, another to load a single product, then another to add it to the cart. The sequence is fixed. It’s predictable, linear, and designed by a developer in advance.
With A2A, we step into a non-deterministic world — where machines talk to each other more like people. One agent might ask another to “find all articles about Donald Trump,” then follow up with “only keep the ones from the past 24 hours,” and then “sort them in reverse chronological order.”
It’s not a sequence of fixed endpoints — it’s a conversation.
That opens the door for publishers to expose their own agents — a “newsroom agent,” a “fact-checker agent,” a “quote search agent” — that can be hired by others on demand.
So how can publishers make money from this?
This isn’t about using AI inside the newsroom. It’s about turning your existing content and talent into AI services that other companies can pay to use. Here are a few real-life examples:
1. Research Agent for Consultants
Who buys it: Strategy firms, analysts, investors
What it does: An AI assistant trained on your archives. It answers questions like “What happened to the EV market after COVID?” — with references, quotes, and summaries.
Why it works: Like a Bloomberg terminal that can talk, cite, and reason with your content.
2. Personalized News Briefings
Who buys it: Risk teams, corporate comms, execs
What it does: Their internal AI asks your agent: “Give me this week’s news on shipping strikes in Asia.” Your agent responds with summaries, charts, or headlines — even in audio if needed.
Why it works: More useful than RSS. Smarter than alerts. Feels like talking to an editor.
3. AI-Powered Fact-Checker
Who buys it: Brands, agencies, internal comms teams
What it does: As companies write blog posts or whitepapers with AI, they call your fact-checking agent to verify claims or suggest sources.
Why it works: Grammarly catches grammar. Your agent catches BS.
4. Crisis Monitor Agent
Who buys it: Consumer brands, public sector, banks
What it does: A real-time media monitoring agent that detects issues (recalls, scandals, protests) and advises the client’s system on how to respond — with facts and historical context.
Why it works: It’s Meltwater, but smarter, faster, and ready to act.
5. News Licensing for AI Outputs
Who buys it: LLM app builders, customer service tools
What it does: Lets other AI agents legally use quotes, facts, or photos from your archive — and tracks what’s used.
Why it works: Like Getty Images, but for verified knowledge snippets.
Will any of this actually take off?
That’s the big question. The protocols are brand new. A2A went live April 9th. MCP is still early. Real adoption will depend on three things:
- Integration speed — Can enterprise tech teams wire this up easily?
- Revenue models — Will it be pay-per-query, per-agent-minute, or licensing bundles? And how easy would it be to pay (would we have a global B2B Agents compensation chambers? A global registry of agents?)
- Rights management — Can publishers protect and meter what’s being used (once given away, how make sure data is not stollen? Who is liable for misconduct of AI agents?)
But one thing is clear: if AI agents are the new browsers, media brands could finally move beyond ads and paywalls — and start getting paid every time a robot needs to think.
Instead of just producing content, you become an intelligence provider — to machines.
And that might just be the most valuable audience of all…
