What is MCP, and what does it mean for your marketing system?

MCP, the Model Context Protocol, is an open standard that lets AI assistants connect to other software and data. It means an assistant like Claude or ChatGPT can read and act on your business systems, within the permissions you set. According to Anthropic, December 2025, there are more than 10,000 active public MCP servers.

What is MCP in plain English?

Think of MCP as a common plug. Before it, every AI tool needed its own custom connection to every app.

The official specification describes MCP as an open protocol that connects AI applications to outside data sources and tools. A business system offers an MCP server, and an AI assistant connects to it.

The specification lists what a server can offer. Tools are actions the AI can run, resources are data it can read, and prompts are ready made instructions for users.

Once connected, the assistant can look things up and, where allowed, take actions. It might find a contact, check a booking or draft an email in your CRM.

Anthropic introduced MCP in late 2024. In December 2025 it donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation set up by Anthropic, Block and OpenAI.

Google, Microsoft, AWS, Cloudflare and Bloomberg support the foundation too, according to Anthropic. No single company controls the standard.

Which AI assistants support MCP?

The main ones. According to Anthropic, December 2025, MCP has been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot, Visual Studio Code and other AI products.

OpenAI built its Apps SDK for ChatGPT on MCP, announced in October 2025. Full MCP connections are currently available to ChatGPT Business, Enterprise and Edu users, according to the OpenAI Help Center in October 2026.

OpenAI says ChatGPT prompts you to connect the first time you use an app, so you know what data may be shared. For write or modify actions, ChatGPT may ask for confirmation, depending on the app's permissions.

Broad support matters. A system you connect once can work with whichever assistant your team prefers, and you are not tied to one AI company.

The key figures on MCP and AI agents

• There are more than 10,000 active public MCP servers, according to Anthropic, December 2025.
• MCP has been adopted by ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code, according to Anthropic, December 2025.
• The MCP software development kits see more than 97 million downloads a month across Python and TypeScript, according to Anthropic, December 2025.
• 40% of respondents from large organisations say they are scaling AI agents, up from 27% a year earlier, according to McKinsey, August 2026.
• About 2 in 10 respondents say they have reached the scaling phase with AI agents across their organisation, according to McKinsey, August 2026.

How is MCP different from an API or an integration?

An API is a door into one app, built for developers. An integration links 2 apps and passes data between them on rules someone set up in advance.

MCP sits on top of those. It describes what a system can do in a way an AI assistant can understand and use when asked.

So instead of building a new report or automation for every question, your team asks. The assistant works out which tools to use, within the permissions it has.

Integrations and automations still matter. Fixed tasks that must run the same way every time, such as a booking reminder, belong in an automation rather than a chat.

What can MCP do for a marketing team?

It lets your team work with the business system in plain language, from the assistant they already use.

Here is what that can look like, with the right permissions in place:

• Ask which enquiries from last week have not had a reply, and draft one for each.
• Pull this month's bookings by source and summarise what changed.
• Find every client who has gone quiet for 90 days and suggest a check in message.
• Draft next week's email campaign from the latest article, ready for approval.
• Add notes from a call to the right contact and set the follow up task.

Each of these used to mean logging in, exporting a report or asking a colleague. The assistant does the legwork, and a person still decides.

It works best for questions and drafts. Anything that goes out to a client should still pass a person first, at least until you trust the setup.

Is MCP safe to connect to your CRM?

It can be, if it is set up with care. The MCP specification says users must explicitly consent to and understand all data access and operations.

It also says hosts must obtain explicit user consent before invoking any tool. MCP cannot enforce these principles itself, so the people building each connection have to.

The official security guidance recommends a least privilege model, where each connection gets only the access it needs. It warns against wildcard or full access permissions.

The OpenAI Help Center warns that untrusted MCP servers may increase exposure to security risks, including prompt injection. Prompt injection is when hidden instructions in content try to steer the assistant.

A safe setup checklist:

• Connect only servers from suppliers you trust.
• Give each connection the narrowest permissions that do the job.
• Start with read only access, then add actions such as sending or deleting one at a time.
• Require approval before anything is sent to a client or changed in bulk.
• Keep a log of what the assistant read and did.
• Review permissions every quarter and remove what nobody uses.

What does MCP mean for your marketing system?

It raises the value of one connected system. An assistant linked to a contact list sees one part of the picture.

An assistant linked to a system holding contacts, campaigns, bookings and payments sees all of it. It can answer questions that cross them, such as which campaign brought in clients who paid.

It also raises the stakes on clean data. If records are duplicated or out of date, the assistant will repeat those mistakes faster.

What to have in place before you connect an assistant:

• One CRM that holds every contact, with duplicates merged.
• Clear names for pipelines, stages, tags and fields, so the assistant reads them correctly.
• A written list of what the assistant may do alone and what needs approval.
• One person who owns the connection and reviews its log.

Agents are spreading quickly. McKinsey ran its survey in May and June 2026. 40% of respondents from large organisations said they are scaling AI agents, up from 27% a year earlier.

Our team connects Claude, ChatGPT and other assistants that support MCP as part of every bespoke marketing system we build. We set the rules for what the assistant can see and do, and we set it up with you on a call.