FigmaMCP

How Figma MCP works, from your request to your file

How Figma MCP works is simple to state. Your agent, Claude Code or Codex, calls tools on a small MCP server on your Mac. A plugin in the Figma desktop app then reads or edits the file you have open and returns the result.

The three parts

The parts of FigmaMCP
Part Where it runs What it does
Your agentClaude Code or CodexReads your request and decides which tool to call.
The MCP serverOn your MacOffers the tools to the agent and passes each call on.
The pluginInside the Figma desktop appReads and edits the file you have open, then returns the result.

The server talks to the plugin, and the plugin acts in the file through Figma’s Plugin API. There is no Figma access token to create or paste.

Download for Mac Version 0.1.10

Figma MCP how it works: one request, step by step

  1. You ask

    For example: “Rename every layer called Frame 12 to match its heading.”

  2. The agent reads first

    It calls page_info and node_info to find the layers and their exact names.

  3. The agent writes

    It calls a write tool and passes each current name back. A wrong name is refused.

  4. The plugin checks and acts

    The plugin checks scope and names. If they pass, it makes the change in your file.

  5. The agent checks

    It can read the layers again, or export a picture with node_export_visual.

Figma MCP what does it do

What FigmaMCP does, with examples
Group Examples
ReadList pages, layers, components, styles and variables. Get a picture of a layer.
CreateFrames, shapes, text, SVG layers, components, variants and instances.
ChangeRename, move, resize, copy, group and delete layers. Set fills, text, effects and auto layout.
Design systemCreate and edit variables and styles. Bind variables to layers.
Notes and prototypesWrite annotations and prototype interactions.

For the full list, see all tools. It does not generate code, and it does not import HTML.

Figma MCP how to use it

  1. Install

    Download the Mac app and follow its checklist. See Figma MCP setup.

  2. Open a file

    Use the Figma desktop app.

  3. Ask

    Ask Claude Code or Codex. Name the page or frame, and say what must stay the same.

  4. Review

    Look at the result. Use Figma version history to go back.

Why it reads before it writes

The file is a live document. You or a teammate may rename or move a layer between two calls. So the agent reads the current names first, and each write passes them back. If a name no longer matches, the plugin refuses the edit, and the agent reads again. This is how a wrong target gets caught before it changes anything.

What keeps it in bounds

Your agent sends what it reads to its own model provider, under that provider’s terms. FigmaMCP sends no design content to us. For background, see what is a Figma MCP server and edit Figma with AI.

Questions

How does Figma MCP work in one sentence?
Your agent calls tools on a local MCP server, and a plugin inside the Figma desktop app reads or edits your open file and returns the result.
What does FigmaMCP actually do?
It reads your file and can create and change layers, components, styles and variables. It does not generate code and it does not import HTML.
Do I need to learn commands to use it?
No. You ask your agent in plain words, and the agent picks the tools.
Does it work outside a Mac, or in the browser?
No. FigmaMCP supports macOS only for now, and it needs the Figma desktop app.