# What is an MCP server? A plain guide for creators > MCP without the jargon — what an MCP server is, how Claude, ChatGPT and code editors use one to call tools, and how eroq's turns a chat into an image and video studio. Published 2026-09-28 · eroq.ai — canonical: https://eroq.ai/blog/what-is-an-mcp-server "Add our MCP server" has become the standard way for software to plug into AI assistants, and the phrase shows up everywhere now — Claude, ChatGPT, Cursor, VS Code. If you make images or video rather than write code, here is what it actually means, why it matters to you, and what changes when a studio like eroq runs one. ## The one-sentence version An **MCP server** is a small service that tells an AI assistant, "here are the things I can do, and here is how to ask me" — and then does them when the assistant asks. MCP stands for **Model Context Protocol**, an open standard started by Anthropic and now supported by most assistants. Before it, every AI product needed its own custom integration with every tool. With it, a tool publishes one MCP server and every assistant that speaks the protocol can use it. ## What an assistant sees When you connect an MCP server, the assistant receives a list of **tools**, each with a name, a description and the inputs it takes. eroq's list looks like this: - `generate_image` — make a picture from a prompt (and, optionally, a reference photo). - `generate_video` — start a video clip on a named engine; `get_video_status` follows it. - `generate_speech` — read a line in a chosen voice. - `enhance_prompt` — turn a rough idea into a detailed prompt. - `list_models`, `list_voices`, `list_characters`, `get_account` — look up what is available and what it costs. The assistant reads these descriptions the way you would read a menu. When you type "make me a vertical clip of rain on a neon street", it picks `generate_video`, fills in the prompt, the 9:16 format and a length, and calls it. You never see the plumbing unless you ask. ## Why creators should care - **Your conversation becomes the brief.** You already explain what you want to an assistant; with an MCP server, that explanation becomes the render instead of a prompt you copy into another tab. - **The assistant is a decent prompt writer.** It turns "moody, rainy, lonely" into subject, light, lens and motion — the vocabulary engines respond to — and eroq's free `enhance_prompt` tightens it further. - **Iteration stays in one place.** "Warmer." "Now vertical." "Same scene, five seconds, slow push-in." Each follow-up is one sentence. - **Nothing new to pay for.** eroq's tools bill the studio's credit prices — 10 credits for a Krea 2 image, 45 for a five-second Seedance 2.0 Mini clip — from the same wallet. ## Remote vs local, and why sign-in matters MCP servers come in two shapes. A **local** server runs on your computer and is started by the app (common for developer tools). A **remote** server lives on the web at a URL — eroq's is `https://eroq.ai/mcp` — and works from anywhere, including the Claude and ChatGPT websites. Remote servers need to know who you are. The old way was to paste an API key somewhere, which is exactly the kind of secret that ends up in a screenshot. The current way is **OAuth**, the same "Sign in with…" flow you know from other apps: the assistant sends you to eroq, you sign in and click Allow, and the assistant gets its own key that you can revoke at any time. eroq supports it for Claude, ChatGPT, Claude Code, Cursor and VS Code. ## Where to connect eroq - **Claude** — one click: the [Connect eroq to Claude](/blog/connect-eroq-to-claude) button opens Claude's connector dialog pre-filled. - **ChatGPT** — a developer-mode app with OAuth; the [ChatGPT guide](/blog/use-eroq-in-chatgpt) walks through it. - **Claude Code, Cursor, VS Code** — one command or one click from [eroq.ai/mcp](/mcp). - **Your creative apps** — Photoshop, Premiere Pro, After Effects, DaVinci Resolve, Figma and Blender get native [plugins](/plugins) rather than MCP, because there the output belongs in your project, not in a chat. ## What an MCP server cannot do It does not make the assistant smarter about images — the render quality is the engine's. It does not bypass anyone's rules: the assistant applies its own usage policy to what it will request, and eroq's [acceptable-use policy](/legal/aup) decides what renders (adults only, fiction only, no real people without consent, never minors). And it does not remove waiting: video still takes one to five minutes, so the assistant starts the clip and checks back. ## FAQ ### Is MCP only for developers? No. Setting one up used to mean editing config files; the one-click and sign-in flows now make it a two-minute task in Claude or ChatGPT. ### Is it safe to connect a server to my assistant? Connect servers you trust, the same as any app you give access to. With OAuth you can see exactly which app holds a key and revoke it — eroq lists each one as « App (connected) » under Developers → API keys. ### Does MCP cost anything? The protocol is free. What you pay for is what the tools do — on eroq, the same credits as the studio, with free lookups and automatic refunds for failed renders. ### Which assistants support MCP? Claude, ChatGPT (developer mode), Claude Code, Cursor, VS Code with Copilot, and a growing list of others. The one-click buttons for each are on [eroq.ai/mcp](/mcp).