What Is MCP? A Publisher's Plain-English Guide to Model Context Protocol

A quiet shift is happening in how AI systems find and use information. Until recently, AI agents had two options: ask a human to paste in content, or scrape the web and hope for the best. Neither option served publishers well.

MCP, short for Model Context Protocol, changes that. It's the protocol that lets AI agents query your content directly, in real time, through a controlled channel you define. Downloads of the MCP software development kit hit 97 million per month by March 2026, up from roughly 100,000 at launch. That is a 970x increase in 18 months. Every major AI company has adopted it.

Publishers who understand MCP now are the ones who will build the commercial infrastructure that controls how their content flows into AI products. Those who don't will find out about it later, on worse terms.

MCP is an open standard that defines how AI systems connect to external data sources, tools, and services. Instead of an AI agent processing only what's in its immediate conversation, MCP gives it a structured way to query live data, search databases, read documents, and call external services in real time.

Think of it this way. Before MCP, connecting an AI agent to a data source required custom engineering for every single integration. MCP standardizes that connection the way USB standardized how devices plug into computers. One protocol, any data source, any AI system.

Anthropic introduced MCP in November 2024 as an open-source standard. By December 2025, it had been donated to the Agentic AI Foundation under the Linux Foundation, with OpenAI, Google, Microsoft, AWS, and Cloudflare as supporting members. It stopped being Anthropic's protocol and became industry infrastructure.

The speed of MCP adoption has no real precedent in developer tooling. OpenAI adopted MCP across its Agents SDK and ChatGPT desktop in March 2025. Google DeepMind confirmed support in April 2025. Microsoft shipped MCP servers for GitHub, Azure, and Microsoft 365 by Q3 2025.

By Q1 2026, independent researchers had indexed over 17,000 MCP servers across public registries. And 67% of CTOs surveyed say MCP will be their default agent-integration standard within 12 months.

Why did it win so fast? Because it solved a real problem cleanly. Before MCP, every team building an AI agent had to write custom connectors for every data source they needed: one for their CMS, one for their database, one for their knowledge base, and so on. MCP made those connectors interoperable. Build once, connect to any MCP-compatible AI system. The incentive to adopt was immediate and obvious for both sides: AI builders got easy access to data, and data owners got a standard way to expose it.

That combination of clear value and open governance made it the only protocol with real cross-industry momentum.

Publishers should care about MCP because it is rapidly becoming the default channel through which AI agents access content. If your content is valuable to AI systems, MCP is the infrastructure layer that determines whether you control that access or not.

Without an MCP server, your content is accessible only through scraping or bulk licensing deals. Both options put AI companies in control. Scraping means no consent, no payment, and no record of what was used. Bulk deals typically favor the AI company: flat fees, broad rights, and no granular tracking.

With an MCP server, you set the terms. You decide which content is queryable. You define the pricing per query. You track exactly what was accessed, by which AI system, at what time. Publishers who expose content via MCP can charge AI agents per query, creating a metered revenue stream rather than a one-time licensing fee.

This is the commercial difference that matters. A scrape produces nothing. A bulk license produces a one-time payment. An MCP endpoint produces revenue every time your content is used, for as long as AI agents keep querying it.

An MCP server is a piece of software that sits between your content and AI systems. When an AI agent needs information, it sends a structured query to your MCP server. The server searches your content, retrieves the relevant material, and sends back a clean, structured response. No scraping. No raw HTML. No lost context.

From the publisher's side, the MCP server enforces your rules. It authenticates each AI system before giving access. It applies your pricing. It logs every query with a timestamp and the identity of the requesting agent. It can expose some content publicly, keep other content behind a paywall, and block specific AI systems entirely.

For a scientific publisher, this means an AI agent can query your full archive of research papers without ever receiving the raw files. The AI gets structured answers to specific questions. You retain control of the underlying content. Every interaction is logged and billed.

This is a fundamentally different relationship than what most publishers have with AI companies today. It is structured, traceable, and commercial from the first query. Learn more about how AI-ready data infrastructure makes this possible at scale.

The distinction between scraping and MCP access is not just technical. It is legal and commercial.

When an AI company scrapes your content, it takes a copy. That copy gets processed, stored, and used in ways you cannot track or enforce against. You have no record of what was taken. You have no way to prove how it was used. And you have no revenue from the interaction. This is the situation the Meta copyright lawsuit exposed: an AI company scraped at massive scale, left no auditable trail, and argued the entire thing qualified as fair use.

When an AI agent queries your MCP server, it receives a structured response to a specific query. It does not receive a copy of your underlying content. The query is logged. The payment is processed. The rights terms are applied at the moment of access.

This distinction is what makes MCP genuinely useful for protecting your content from AI scraping. It's not a legal defense strategy. It's an infrastructure strategy. If your content is only accessible through a channel you control, unauthorized access becomes technically harder as well as legally riskier.

The IAB Tech Lab has formed a dedicated AI Content Monetization working group to standardize exactly these commercial interactions between publishers and AI systems. MCP is the technical layer that makes those commercial standards enforceable.

The most direct path for publishers is to deploy a dedicated MCP server for their content archive. This exposes your catalog to AI systems through a controlled, authenticated channel, with usage metering built in.

Several paths exist depending on your technical resources and commercial goals.

For publishers with engineering teams, building an MCP server on your own infrastructure gives you the most control. You decide exactly what data the server exposes, which authentication methods to require, and how to structure pricing. Your content never leaves your servers.

For publishers without dedicated engineering resources, managed platforms can deploy an MCP endpoint on your behalf. TollBit operates as a marketplace where publishers can set per-query pricing and let AI agents discover and pay for their content through a shared infrastructure layer.

The key question to ask before deploying is not just "can AI agents access my content?" but "can I see exactly how my content is being used, by whom, and for what purpose?" That traceability is what transforms an MCP endpoint from a simple access channel into a full content monetization model.

Publishers who need to meet EU data residency requirements, or who cannot route their content through third-party infrastructure, need a deployment model where the MCP server runs on their own infrastructure. That is a requirement that managed platforms cannot meet. It is the reason regulated publishers and national institutions require dedicated infrastructure rather than shared endpoints.

MCP is not a trend to watch. It is infrastructure that is already being built around your content, with or without your participation. The question is whether you are inside that infrastructure or outside it.

Ninety-seven million monthly SDK downloads. Over 17,000 servers indexed. Adoption by every major AI company. MCP has crossed the threshold from emerging standard to default protocol in under two years.

Publishers who deploy their own MCP endpoints now will control how AI systems access their content for years to come. They'll set the pricing, track the usage, and build commercial relationships with AI platforms on their own terms. Publishers who wait will negotiate from a weaker position, or find themselves defending their rights in court.

The infrastructure for protecting and monetizing your content for AI starts with understanding MCP. Then it starts with deploying it.

Talk to Alien Intelligence about building your MCP infrastructure.

MCP stands for Model Context Protocol. It was created by Anthropic and released as an open-source standard in November 2024. In December 2025, Anthropic donated the protocol to the Agentic AI Foundation under the Linux Foundation, with OpenAI, Google, Microsoft, AWS, and Cloudflare as co-sponsors. It is now governed as shared industry infrastructure rather than a proprietary Anthropic standard.

A regular API is a custom connection built between two specific systems. MCP is a universal standard that any AI agent can use to connect to any MCP-compatible data source. The difference is like the difference between a custom cable and a USB port: APIs require custom engineering for every integration, while MCP lets any compatible AI system connect to any compatible data source without additional engineering. This is why it spread so quickly once the major AI platforms adopted it.

Yes. One of MCP's core features for publishers is the ability to attach pricing to queries. When an AI agent queries your MCP server, you can require payment per query before returning any content. Platforms like TollBit have built marketplaces on top of this capability, letting publishers set per-query prices and letting AI companies discover and access licensed content at scale. Publishers who run their own MCP infrastructure can implement any pricing model they choose, including per-query fees, subscription access, or tiered pricing by content type.

It depends on the deployment model. Building your own MCP server on your own infrastructure requires engineering resources, but gives you full control over data residency, pricing, and access rules. Managed platforms can deploy an MCP endpoint on a publisher's behalf with minimal technical setup, but those solutions use shared infrastructure that may not meet EU data residency requirements or the sovereignty needs of regulated institutions. Publishers with strict data requirements — national archives, scientific publishers, regulated broadcasters — typically need the full on-premises deployment model.

No. They are different layers of the same problem. Cloudflare Pay-per-Crawl charges AI bots when they scrape raw pages from a website, using Cloudflare's CDN infrastructure to intercept and bill the request. MCP replaces the crawl entirely: instead of letting bots scrape raw HTML, an MCP server delivers structured, rights-cleared responses to specific queries. MCP gives publishers more control, better traceability, and the ability to serve structured data rather than raw page content. The two approaches can coexist: Pay-per-Crawl to handle bots hitting the public website, MCP to serve premium structured access to licensed content.