# What is MCP? Easy explanation of the standard connecting AI to your data and tools

URL: https://opcaios.org/en/blog/mcp-la-gi
Date: 2026-10-22
Author: Lê Quốc Nghĩa

What is MCP? MCP ([Model Context Protocol](/tien-ich/model-context-protocol)) is an open standard that helps AI applications like [Claude](/tien-ich/claude), [ChatGPT](/tien-ich/chatgpt), or [Cursor](/tien-ich/cursor) connect to external data and tools — files on your computer, Google Drive, CRM, websites, databases — in a common way. Instead of each AI application having to write separate integrations for each piece of software, the software only needs to build one "MCP server" and every AI application that supports MCP can use it. This post explains MCP in everyday language, how it works, how to get started, and the risks you need to know before connecting AI to your business data.

![Diagram of MCP: AI application connecting through MCP to files, Google Drive, CRM and websites](https://opcaios.org/images/blog/mcp-la-gi/so-do-mcp.webp)
*An AI application (host) opens multiple MCP connections, each one to an MCP server. Diagram: OPCAIOS.*

## What is MCP?

MCP (Model Context Protocol) is an open standard that helps AI applications connect to external systems in a common way, and the official documentation compares MCP to **a USB-C port for AI applications**. USB-C is a common standard port for all kinds of devices, while MCP is a common standard between AI applications and data or tools.

Before MCP existed, to get an AI assistant to read sales data, someone had to write an integration specifically between one particular AI application and one particular sales software, then do it all over again when you switched applications. With MCP, sales software only needs one MCP server, and Claude, ChatGPT, VS Code, Cursor can all connect to it. Anthropic announced MCP at the end of 2024 as open source (Anthropic), and the current specification version is **2026-07-28** according to the MCP specification page. MCP names versions by date when breaking changes occur, and is supported by Microsoft's Visual Studio Code, Cursor, Claude, and OpenAI's ChatGPT.

## What can MCP help AI do?

MCP helps AI read data and take actions in the systems you already use. There are three examples, from the MCP introduction page: an AI assistant reads your Google Calendar and Notion notes, [Claude Code](/tien-ich/claude-code) builds an entire web app from a [Figma](/tien-ich/figma) design, and an internal chatbot connects multiple databases so employees can ask for data in plain language.

For a small business in Vietnam, closer examples include three tasks:

- Ask "how many customers submitted forms on the website this week?" and AI reads it from your system instead of you opening each page.
- Have AI write a draft blog post and automatically push it straight to your website admin panel, waiting for you to review.
- Let AI read your contracts folder on your computer to summarize terms, without having to upload each file into the chat.

These three tasks are example illustrations from OPCAIOS, not survey data.

## How MCP Works

MCP works according to a three-role model: host, client, and server. A host can connect to multiple servers at once, and each connection is a separate client, according to the MCP introduction page.

| Role | What it is | Examples |
|---|---|---|
| **Host** | The AI application you are using | Claude Desktop, Claude Code, ChatGPT, VS Code, Cursor |
| **Client** | A connection inside the host, each client connects to one server | The host creates it, you don't have to do anything |
| **Server** | A program that provides data and tools to AI | Server reads files, Google Drive server, CRM server, website server |

For example, Anthropic's Claude Code connects to a GitHub server, a database server, and a website server all at once, and each connection is a separate client. Under the hood, all messages follow the JSON-RPC 2.0 format, but you don't need to know this if you're just using it, because the official SDKs handle that part for the person writing the server.

### What three things does an MCP server provide?

An MCP server provides three things: tools, resources, and prompts. These three things are, according to the MCP introduction page:

1. **Tools:** actions that AI can call, such as creating leads, sending emails, querying a database, writing draft articles.
2. **Resources:** data for AI to read as context, such as file contents, database records, API results.
3. **Prompts:** pre-written request templates that help users call the right workflow without having to write it themselves.

In practice, **tools** are the most used part. For example, Claude reads the list of tools the server declares (name, description, parameters), decides which tool to call, and uses the result to answer you.

### Where does an MCP server run: on your machine or on the network?

An MCP server runs in two ways: locally on your machine or remotely on the Internet. These two ways differ in connection method, according to the MCP introduction page:

- **Local server (stdio):** runs right on your machine, the host starts it as a child process. Good for reading files on your machine and controlling applications on your machine.
- **Remote server (Streamable HTTP):** runs on the Internet, the host connects via HTTP, usually logging in with OAuth or an API key. Good for online services like Sentry, Notion, or your own website.

For example, a server that reads a contracts folder should run locally, while a Notion or Sentry server runs remotely.

## How to Start Using MCP?

### In chat applications (Claude, ChatGPT)

Large AI assistants call MCP servers from a distance as "connectors". You go to the connector settings section, choose a service from the list or paste the MCP server address, log in, then turn it on in the conversation. You don't need to install anything on your machine.

### In Claude Code (command line)

[Claude Code](https://opcaios.org/tien-ich/claude-code) adds servers using the `claude mcp add` command:

```bash
# Remote server via HTTP
claude mcp add --transport http sentry https://mcp.sentry.dev/mcp

# Local server running via npx, pass key through environment variable
claude mcp add my-server -e API_KEY=xxx -- npx my-mcp-server

# See the servers you've added
claude mcp list
```

By default the server only works in the current project; add `--scope user` to use it in all projects, or `--scope project` to share the configuration with your team via the `.mcp.json` file.

### Test a server before using it

[MCP Inspector](https://modelcontextprotocol.io/docs/tools/inspector) is the official tool to open a server, see what tools and resources it declares, and try calling each tool. You should test it with Inspector before plugging an unfamiliar server into an AI application that contains real data.

## Real example: What can the opcaios.org MCP server do?

The opcaios.org MCP server is a dedicated server for AI assistants to work with the content and customers of this website itself. The server declares three groups of tools, according to the website's admin (checked on 22/10/2026):

- **Content:** `list_posts`, `get_post`, `check_post_seo`, `create_post`, `update_post`, `get_brand_voice` (brand voice).
- **Customers and sales:** `list_submissions` (forms customers leave), `list_contacts`, `create_lead`, `list_deals`.
- **Shortened links:** `create_link`, `get_link_analytics`.

The important point is permission limits. Each connection is given a key with separate permissions, **read-only** or **read and write**, so the assistant can only query without accidentally modifying data. Also the admin page has a **Review mode**: when turned on, all posts the AI creates via MCP are only saved as drafts and put in a queue, posted to the website when the reviewer clicks approve.

## What risks should I know about before using MCP?

The main risk of MCP is that AI gets permission to touch real data and systems, so you need to be careful like when giving access to a new employee. The official security documentation emphasizes five points, according to the MCP introduction page and Anthropic:

- **Check the source before installing.** Local servers run with your own permissions: they can read files and run commands. Only install servers from trusted developers.
- **Treat tool descriptions as untrusted data.** A malicious server can hide instructions in tool descriptions, and content that AI reads about can also contain injected commands, called prompt injection.
- **Grant minimum permissions.** Start with read-only access, limit directories, give each server its own API key.
- **Keep an approval step for irreversible actions:** sending email, deleting data, transferring money, posting publicly.
- **Run in isolation when possible.** Unknown servers should run in a container or virtual machine.

## Where should I start using MCP?

Start with a recurring weekly task that needs data from one system, with read-only permission. Four steps in order:

1. **Pick a recurring weekly task**, for example summarizing form responses or looking up orders.
2. **Find an existing MCP server** for that system (many large services already have official servers; see [MCP Registry](https://opcaios.org/tien-ich/mcp-registry) or the [MCP Servers](https://opcaios.org/tien-ich/mcp-servers) list).
3. **Connect with read-only permission**, try it for 1 week.
4. **Only enable write permission** when you trust the results, and always include an approval step.

If you want to write your own MCP server, the free course from [Hugging Face](/tien-ich/hugging-face) is a good starting point, see the post [MCP Course Hugging Face: a guide from A–Z](https://opcaios.org/blog/mcp-course-hugging-face). A summary of the protocol is on the [Model Context Protocol](https://opcaios.org/tien-ich/model-context-protocol) page in the Utilities library.

## Frequently asked questions

### Is MCP free?
Yes. MCP is an open standard, the specification and official SDKs are all open source. Any cost comes from the AI application you use or the service behind the MCP server, such as Claude's plan, ChatGPT, or a CRM service. Since pricing changes over time, check the official pricing page of each service; this post doesn't give specific numbers because OPCAIOS hasn't verified the price tables. For example, a server that reads files on your machine costs nothing extra beyond the AI application you're using.

### Can MCP only be used with Claude?
No. MCP was started by Anthropic but is an open standard, so OpenAI's ChatGPT, Microsoft's Visual Studio Code, Cursor, and many other applications support it. One MCP server written once can be used with any AI application that supports this standard, according to the MCP introduction page. Ultimately, specific support from each application changes by version, so check the application's documentation before installing.

### How is MCP different from an API?
An API is a private door for each software, each one works differently, while MCP is a standardization layer on top. An MCP server usually calls the API of the software underneath, but presents it again in a common format that every AI application supporting MCP understands, with descriptions so AI knows when to use which tool. For example, opcaios.org's MCP server calls internal APIs to create a draft post, but the user just sends one message to Claude. See also the post [What is an API](https://opcaios.org/blog/api-la-gi).

### Can I use MCP if I don't know how to code?
Yes, if you use a connector already built into Claude or ChatGPT: just turn it on and log in. Writing your own MCP server or installing a local server requires a little command-line knowledge. Start by trying a read-only connector, then get used to seeing what tools AI calls, and only then think about installing a local server. If you want to learn properly for a small team, see [AI application training course](/services/dao-tao-ung-dung-ai).

### Is using MCP safe?
As safe as you are careful about granting permission. Only install servers from trusted sources, according to the MCP introduction page, start with read-only permission, and keep an approval step for every action that can't be undone like sending email, deleting data, or posting publicly. For example, on opcaios.org, Review mode keeps every post created by AI in draft status until someone clicks approve.

*Source: [official Model Context Protocol documentation](https://modelcontextprotocol.io) (spec 2026-07-28: introduction, architecture, local server security), `claude mcp add --help` command. Updated 22/10/2026.*
