What is an MCP (Model Context Protocol)?

An MCP (Model Context Protocol) is an open standard that connects AI models to external tools, data and applications. Created by Anthropic in November 2024 and donated to the Linux Foundation in December 2025, it is now supported by ChatGPT, Gemini, Copilot and Claude, with more than 10,000 public servers available.

Think of MCP as the USB-C of AI: one universal plug. Before it, every connection between a model and a tool (your CRM, your database, your email) required a custom integration. With MCP, you build a server once and any compatible AI assistant can use it.

How does an MCP work in practice?

The protocol has two sides. On one side sits the client: the AI application the user works with, such as Claude, ChatGPT or Copilot. On the other sits the MCP server: a program that exposes the capabilities of an external system in a format the model understands.

An MCP server can expose three things: tools (actions the model can execute, like “create a contact in the CRM”), resources (data the model can read, like reports or documentation) and prompts (reusable instructions). When you ask your assistant “how many leads came in this week?”, the model calls the right tool on the server, gets the data back and answers with real numbers instead of guesses.

Why did MCP become the standard?

Because it solved the integration explosion, and because competitors adopted it instead of building rivals. OpenAI announced support in 2025, followed by Google and Microsoft. In December 2025, Anthropic donated the protocol to the Agentic AI Foundation, a Linux Foundation directed fund co-founded with Block and OpenAI and backed by Google, Microsoft, AWS, Cloudflare and Bloomberg. It stopped being one vendor’s technology and became neutral industry infrastructure.

The numbers back this up: more than 10,000 active public MCP servers by late 2025, according to Anthropic, and 41% of software organizations running MCP servers in limited or broad production, according to Stacklok’s 2026 report.

What changes for marketers?

MCP turns your AI assistant into a colleague that works inside your tools. Concrete examples:

• CRM and sales: an MCP server connected to HubSpot or Salesforce lets the model query contacts, segment lists and prepare personalized outreach with real data.
• Analytics: connected to GA4 or your data warehouse, the assistant answers business questions (“what was campaign X’s CPL?”) without manual exports.
• Content and publishing: CMS servers let you draft posts, update pages and manage campaigns directly from the conversation.
• Commerce: as AI agents start researching and buying on users’ behalf, brands with MCP integrations become reachable to those agents first.

There is also an indirect effect: if AI agents will interact with your brand through data and tools, your footprint in AI systems (AIO) is no longer just content. It now includes infrastructure.

What are the risks and limitations?

The ecosystem is young and it shows. Users report installation failure rates of 30 to 50% on community-built servers, and security researchers filed more than 30 CVEs tied to MCP servers in January and February 2026 alone. The practical rules: use official or audited servers, grant the minimum permissions needed, and treat any MCP server the way you would treat a plugin with access to your data.

How do I start using MCPs?

You do not need to code to get started. The typical path for a marketing team:

• Check whether your tools already offer an official MCP server (HubSpot, Notion, Slack, Google and hundreds of others already do).
• Connect those servers to the AI assistant your team uses, through its built-in connector directory.
• Start with read-only use cases (querying data, generating reports) before granting write permissions.
• If you own a product, consider building an MCP server: it makes your platform usable by AI agents.

Next step: learn how your brand enters AI systems in the AI Optimization pillar, and see how to rank on ChatGPT, where AI agents already shape discovery.