What Is MCP (Model Context Protocol)? A Plain Guide
MCP, the Model Context Protocol, is an open standard that lets AI apps like Claude and ChatGPT connect to outside tools and data in one common way. A service builds one MCP server, and any AI app that speaks MCP can use it.
That might mean searching your notes or reading a database. The official docs compare it to a USB-C port for AI applications.
What problem does MCP solve?
Before MCP, every AI app had its own way of plugging in tools. If you made a calendar integration, you built it once for one chatbot, again for another, and again for a code editor. Most services didn’t bother, so your AI stayed stuck inside its chat window, knowing only what you pasted in.
MCP replaces those one-off integrations with a shared protocol. The service side writes a server once. The AI side writes a client once. Anything that speaks the protocol on one end can talk to anything on the other.
For you, the result is practical. Your assistant can look things up and save things where your work already lives. You stop copying and pasting.
Who made MCP, and who controls it now?
Anthropic introduced MCP in November 2024 as an open, open-source standard. In December 2025, Anthropic donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation co-founded with Block and OpenAI. The project’s maintainers still run its technical direction.
That governance matters if you’re deciding whether to build on it. MCP is not tied to one AI company, and the docs list support across Claude, ChatGPT, Visual Studio Code and Cursor, among others.
How does MCP work?
The architecture overview names three participants:
| Part | What it is | Example |
|---|---|---|
| Host | The AI app you use | Claude Desktop, ChatGPT, Cursor, VS Code |
| Client | A connector inside the host, one per server | The piece of Cursor that talks to your notes server |
| Server | A program that offers tools and data | A notes vault, a database, a ticketing system |
A server can offer three kinds of things, which the spec calls primitives:
- Tools are actions the AI can call, like “search notes” or “create issue.” Each tool comes with a description and an input schema, so the model knows when and how to use it.
- Resources are data the app can read for context, like a file’s contents or a database schema.
- Prompts are reusable templates a server can offer, like a standard way to start a code review.
Tools are the part you’ll meet first. When you ask Claude a question and it decides your notes might help, it calls the notes server’s search tool, gets results back, and writes its answer from them.
Under the hood, messages are JSON-RPC. Servers run either locally on your machine (the stdio transport) or remotely over the web (the Streamable HTTP transport). You rarely need to care which, except for setup.
What can you do with MCP?
Here are a few everyday uses, phrased as examples rather than promises about any one product:
- Notes and memory. Your AI saves decisions and ideas to a notes vault, and any other AI connected to the same vault can find them later. More on that in MCP server for notes.
- Docs and workspaces. Notion runs a hosted MCP server so clients like Claude Code and Cursor can read and update pages you have access to.
- Code and developer tools. An editor connects to your error tracker or database so the AI can look at real data while it writes code.
- Files on your computer. A local server gives a desktop AI app access to a folder you choose.
The pattern is the same each time: the AI keeps its own skills, and MCP hands it the right context and actions for the job.
Is MCP the same as an API or a plugin?
Close, but no. An API is a way for programs to talk to one service, and every API is shaped differently. MCP is a standard layer on top: a server usually wraps an existing API and describes its actions in a form any AI client understands.
Older chatbot plugins worked in one product only. An MCP server works in every host that supports the protocol. That’s the main difference you’ll feel: connect a service once per app, and the same server works in Claude today and in Cursor tomorrow.
Is MCP safe to use?
MCP gives an AI real access, so treat a server like any app you sign in to. A few habits help:
- Connect only servers you trust, from the service itself or a source you can check.
- Read what access you’re granting. Remote servers usually sign you in with OAuth and show a consent screen.
- Prefer servers that limit scope. A connector that opens one folder or one vault is safer than one that opens everything.
- Remember the AI sees what the tool returns. If a server returns a private document, that text goes to the model.
The MCP docs also keep a security best practices guide aimed at people who build servers.
How do you connect an MCP server?
It depends on the app. Chat apps such as Claude and ChatGPT add remote servers where you paste a URL and sign in: Claude under Settings → Connectors, ChatGPT under Plugins (older versions: Settings → Connectors). Code editors such as Cursor and VS Code read a small JSON config file with the server’s URL and, often, an API key. The step-by-step setup for Valta shows both styles if you want a worked example.
FAQ
What does MCP stand for?
It stands for Model Context Protocol. “Context” is the point: it’s a standard way to give a model the context it needs from outside the chat.
Do I need to code to use MCP?
No. Using an existing server is a settings task: add a connector and sign in. Building your own server takes code, and the official SDKs cover several languages.
Does ChatGPT support MCP?
ChatGPT does support it. OpenAI’s developer docs cover MCP, and ChatGPT can use remote MCP servers through connectors. Exact availability depends on your plan and settings, so check ChatGPT’s current connector docs.
Is MCP only for Claude?
Anthropic started it, but it’s now governed under the Linux Foundation, and hosts such as ChatGPT and VS Code support it too.
Where Valta fits
Valta is a remote MCP server for Markdown notes. Connect it to Claude, ChatGPT or any other MCP client, and each one can save and search notes in the same vault. What you work out in one AI is there in the next. Valta runs no AI model of its own: your assistant does the writing, and Valta stores the notes and suggests links between them, so they grow into a knowledge graph. If you’ve wondered how Claude and ChatGPT could share memory, this is the mechanism.
There’s a free plan with one vault and 100 notes. Create your vault and connect your first AI in a few minutes.