Black Friday sale: 30% off Flywrite

MCP, short for Model Context Protocol, is an open standard that lets an AI agent such as Claude Code connect to your other apps and data, like Notion or your Google Calendar, so it can look things up and get tasks done there for you.

What it means in practice

On its own, an agent like Claude Code works with the files in your folder. MCP lets it reach the apps you already use. Once an app is connected, you ask in plain words, as you would for any other task. You could ask your agent to:

  • Turn your Notion notes into a first draft.
  • Check your Google Calendar before it plans a publishing schedule.
  • Create Gmail drafts, such as one that sends an issue to your editor.

The MCP site compares it to a USB-C port: one standard plug, so a connection built once works with any agent that supports MCP. Each connection is an MCP server, a small program your agent can call on. Notion, for one, runs its own server, which you add with a single command (below).

Three things trip people up:

  • A server is a program you let your agent use. Connect only servers you trust. Claude Code's docs warn that servers which fetch outside content can expose you to prompt injection: text written to steer the agent.
  • Undo stops at your folder. A checkpoint or a Git commit saves the files in your project. If a tool changed a Notion page or a calendar event, restoring the folder doesn't bring that back.
  • You don't need MCP to write with an agent. A CLI agent reads and edits the files in your folder on its own. MCP is for reaching things outside it. For instructions you reuse, such as a pre-send checklist, agents have Agent Skills instead, which Claude Code runs as skills.

What it looks like

You add a server to your agent with one command. These three come from each agent's own docs:

claude mcp add --transport http notion https://mcp.notion.com/mcp
codex mcp add context7 -- npx -y @upstash/context7-mcp
agent mcp list

The first connects Claude Code to Notion's server. The second adds Context7, a documentation server, to Codex. The third lists the servers Cursor Agent can use. Each agent keeps its own list: Claude Code in ~/.claude.json or a project's .mcp.json, Codex in ~/.codex/config.toml, and Cursor Agent in the same mcp.json as the Cursor editor. A server added to one agent isn't in the others.

How it works

Part What it is
Host The AI app you use, such as Claude Code
Client The connection the host opens to one server, one per server
Server The program that offers tools, resources or prompts

A server can offer three kinds of things:

  • Tools the agent uses to act, such as a search.
  • Resources it reads for context, such as a file or a database.
  • Prompts, ready-made templates for a task.

A local server runs on your computer and talks to the agent over stdio. A remote server runs on the web and connects over HTTP; MCP recommends OAuth to sign in to it. Claude Code lets you choose where a server applies: local scope (the default: this project, only you), project scope (a .mcp.json file in the project folder, shared with everyone who works in it) or user scope (every project, only you).

Anthropic released MCP as open source in November 2024. In December 2025 it donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation that Anthropic co-founded with Block and OpenAI. Claude, ChatGPT, VS Code and Cursor are among the apps that support it.

In Flywrite

Flywrite has its own MCP server. Cursor Agent's one-time setup in Settings → Agents adds it, so Cursor Agent can see the file and lines you have open; approve the Flywrite server the first time Cursor asks. It does nothing in terminals outside Flywrite. Claude Code sees the open file and selection with no setup. The servers you've added to your agent work in Flywrite's terminal too, since it's the same shell as your Mac's Terminal app. More in Writing with Cursor.

Questions

How is MCP different from an API?

An API is one service's own interface, and each app has to be built to use it. MCP is one shared way for an agent to connect to many tools and data sources, so a server built once works with any agent that supports MCP.

Do I need MCP to write with Claude Code or Codex?

No. The agent reads and edits the files in your folder without it. MCP adds things outside the folder, such as your calendar or your Notion pages.

Write with your agent in a writing app.

Buy now