What Is MCP? A Plain Guide for Creators and Agencies

MCP, the Model Context Protocol, is an open standard that lets an AI assistant such as Claude or ChatGPT connect to outside tools and data. Anthropic released it in November 2024, and it now sits with a Linux Foundation fund that OpenAI co-founded. For a creator or an agency, it means you can ask an assistant to run a tool you already pay for, instead of clicking through it yourself.

Vendor pages were read on 5 October 2026. Features change, so check the linked pages before you set anything up.

What is MCP in plain English

An AI assistant on its own can only talk. MCP gives it a standard way to reach other software: read a file, search a database, or press a button inside an app. The official site, modelcontextprotocol.io, describes it this way:

Think of MCP like a USB-C port for AI applications.

Before MCP, every assistant needed its own custom connection to every app. With MCP, an app builds one connection and any assistant that speaks MCP can use it.

Who created MCP and when

Anthropic open-sourced MCP on 25 November 2024. The launch included the specification, software kits for developers, local server support in the Claude Desktop apps, and a set of ready-made servers for tools such as Google Drive, Slack and GitHub.

A year later, on 9 December 2025, Anthropic donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation. Anthropic, Block and OpenAI co-founded it. That matters for you in one practical way: MCP is no longer one company’s product. It is a shared standard that competing assistants support.

How MCP works: the parts

These definitions come from the MCP architecture overview.

Part	What it is	Plain example
Host	The AI app you type into	Claude Desktop, Claude Code, Visual Studio Code
Client	The piece inside the host that holds one connection to one server	One client per connected tool
Server	A program that offers a tool’s abilities to the assistant	A Google Drive server, a GitHub server
Tools	Actions the assistant can take	Search flights, create a calendar event
Resources	Read-only information the assistant can look at	A file, a database schema
Prompts	Ready-made instruction templates	“Summarize my meetings”

The host creates one client for each server you connect. So if you connect three tools, your assistant holds three separate connections.

Tools are the part that does real work. The MCP server guide explains that the model decides when to call a tool based on what you asked. It also says tools may require your consent before they run.

Local and remote MCP servers

A server can run on your own computer or on the internet. The MCP docs call these “local” and “remote”.

• Local servers run on the same machine as the assistant. They suit things like your own files. They usually serve one assistant.
• Remote servers are hosted online, usually by the company that makes the tool. You connect with a web address and usually sign in. They can serve many users at once.

The difference matters for which assistants you can use. Claude’s custom connectors, for example, connect to remote servers from Anthropic’s cloud, so the server must be reachable over the public internet. Local servers configured in Claude Desktop are a separate setup. ChatGPT, according to OpenAI’s help center, connects to remote MCP servers and not directly to local ones.

Which AI assistants support MCP (October 2026)

Support is broad, but the details differ by product and plan. This table lists only what each vendor states in its own documentation.

Assistant	MCP support, per the vendor	Plan notes
Claude (web, Desktop, mobile, Cowork)	Custom connectors using remote MCP	Free, Pro, Max, Team and Enterprise. Free is limited to one custom connector
Claude Desktop	Local MCP servers through a config file	Desktop app on macOS and Windows
ChatGPT	Developer mode with MCP apps; remote servers only	Varies by plan; OpenAI’s pages differ, so check yours
Visual Studio Code (GitHub Copilot chat)	Local and remote MCP servers	Generally available since VS Code 1.102
Gemini CLI (Google)	Local and remote MCP servers	Command-line tool
Microsoft Copilot Studio	MCP tools and resources	For building agents

Sources: Claude custom connectors, MCP local server guide, ChatGPT developer mode, VS Code, Gemini CLI, Copilot Studio.

On ChatGPT, OpenAI’s own pages describe plan access in slightly different ways, and the setup lives behind admin and developer settings. Check your plan’s current page before you count on write actions.

MCP security basics

Connecting a server gives the assistant real power inside another app. Treat it like giving someone a login. Claude’s and ChatGPT’s help pages give the same core advice.

Only connect servers you trust. Claude’s help center says to connect only servers “built and hosted by organizations and applications you trust”. OpenAI says the same: “Only connect servers you trust.”

Read the permissions. When you sign in to a server, look at what it asks to do. Anthropic advises limiting the scopes and denying access if a request seems unnecessary.

Approve tool calls, at least at first. The MCP specification says there should always be a human in the loop who can deny a tool call. Claude’s connector docs add: only click “Always allow” for trusted servers.

Watch for prompt injection. A malicious server can hide instructions that try to make the assistant do something you did not ask for. Both Anthropic and OpenAI name this risk.

Be careful with local servers. The VS Code docs warn that local MCP servers can run arbitrary code on your machine. Install them only from sources you trust.

Turn off what you do not use. In Claude you can switch connectors on or off per conversation. Fewer active tools means fewer surprises.

MCP vs API vs CLI vs agent skills

These four terms come up together, and they do overlap. Here is how they differ.

	What it is	Who uses it	Good for
API	A programming interface a service exposes for code to call	Developers writing software	Custom apps and integrations
MCP server	A standard wrapper that lets any MCP assistant discover and call a service’s actions	Anyone with an MCP-capable assistant	Running a tool by chatting with an assistant
CLI	A command-line program you run in a terminal	Developers, and coding agents such as Claude Code	Scripts, batch jobs, automation
Agent skills	Folders of instructions and scripts an assistant loads when relevant	Assistants that support skills	Teaching the assistant a repeatable workflow

For most services, including FansCreator, an API sits underneath, and MCP and a CLI are two doors into it: one for chat assistants, one for the terminal. Agent skills are different again. Anthropic’s Agent Skills documentation describes them as reusable folders with a SKILL.md file of instructions that load on demand. A skill does not connect to anything by itself. It tells the assistant how to do a task, often using an MCP server or CLI to carry it out.

What MCP means for creators and agencies

For most creators, MCP is useful for one reason: repetitive work. Making 30 variations, upscaling the best 10, and downloading them in the right format is a lot of clicking. An assistant can do that clicking if the tool has an MCP server or CLI.

FansCreator has an API, an MCP server and a CLI, plus agent skills. Through them an AI assistant can use the same features you use in the app: image and video generation, face swap on pictures and video, character swap, editing, upscaling, and downloads formatted for OnlyFans, Fanvue, Instagram, TikTok and X. They are included on FansCreator’s Basic, Pro and Enterprise plans, not on Starter. The connection details come from your FansCreator account.

Who does what

There is one important split. Anthropic’s Usage Policy has a section titled “Do Not Generate Sexually Explicit Content”. OpenAI’s usage policies prohibit non-consensual intimate content and using someone’s likeness without consent.

So in this setup the assistant is the coordinator, not the creator. It writes prompts, queues batches, tracks results and downloads files. The images and video are generated on FansCreator. Keep what you ask the assistant SFW. Adult work belongs in the FansCreator app itself. For the image side of this, see our guide on whether Claude can generate images.

Example workflows (SFW)

These are the kinds of requests that suit an assistant connected to FansCreator. Wording is yours; the assistant picks the matching tools.

Weekly promo set for social. “For my persona, generate 12 casual outfit pictures for Instagram: café, park, gym. Upscale the four best and download them formatted for Instagram.”

Agency batch across creators. “For each of my creators, make three SFW teaser pictures in their usual style, then download each one formatted for Instagram.”

Picture to short clip. “Take yesterday’s beach picture and turn it into a short image-to-video clip with gentle camera movement.” If you are comparing video models first, our Seedance 2.0 guide covers one of them.

Face swap onto a generated clip. “Put my persona’s face on the gym clip you generated this morning.” The video face swap feature handles clips up to 30 seconds, one person per video, and keeps the original audio. Use only your own face, your persona’s, or a face whose owner gave provable consent, and check the platform rule for where you post.

Keep one identity. Each request above names an AI creator you already built, so that creator’s face and body stay the same. You build that creator once, from scratch or from your own reference photos, with the AI influencer generator.

What the assistant does not do

FansCreator does not post or schedule for you, and neither does the MCP server. You download the files and upload them yourself. That keeps the final check with you.

Two rules still apply to every file, however it was made:

• Consent. Only use your own face, your own AI persona, or a face whose owner gave provable consent. The FansCreator Terms of Use prohibit anything else.
• Platform rules. OnlyFans, Fanvue, MYM and Fansly each have their own AI rules. OnlyFans allows AI content only if it clearly features the verified creator and carries a label such as #ai, and its Help Centre lists content that is “wholly AI-generated” or “features heavily altered images” among what it removes. A persona built from scratch does not meet that rule, so it cannot go on OnlyFans. Fanvue and MYM allow labelled AI creators. Fansly bans photorealistic AI even when it is labelled.

A simple way to start

1. Pick one repetitive task you do every week.
2. Connect FansCreator to your assistant with the details from your account.
3. Run a small batch with tool approval switched on.
4. Check the results, then grow the batch size once you trust the flow.

Each generation uses your plan’s credits, so a small first batch also protects your balance.

MCP for creators, in short

MCP is a shared plug that lets AI assistants use other software. Anthropic created it in 2024, the Linux Foundation now hosts it, and Claude, ChatGPT, VS Code, Gemini CLI and Copilot Studio all support it in some form. Connect only servers you trust, read the permissions, and approve actions until you know how a server behaves.

For creators, the payoff is less clicking. You can create a FansCreator account and start on the free plan; MCP and CLI access come with FansCreator’s Basic, Pro and Enterprise plans.