What is figma-pilot?
figma-pilot is an MCP server that lets an AI coding agent control Figma through code execution. Instead of dozens of separate tools it offers three: figma_status, figma_execute and figma_get_api_docs. The agent writes JavaScript against the Figma API and runs it in your open file.
It is an MCP server where the agent controls Figma by writing code. The design follows Anthropic's code execution with MCP approach: a small tool surface, with the agent doing the work in JavaScript rather than calling many narrow tools.
Why only three tools
Every tool definition costs context. figma-pilot ships figma_status, figma_execute and figma_get_api_docs, and the README reports 90 percent or more fewer tokens in tool definitions than a server with dozens of tools.
- Batch operations let one call change a hundred elements.
- Data can be filtered in code before it returns to the model.
- Loops, conditionals and error handling work inside a single call.
What it was used for
The README demos include building an OpenAI-style landing page and generating a design system's components from a screenshot, both inside Figma. For other routes into Figma, see figma-ui-mcp and Figma Unified MCP.
Requirements
- Node.js 18 or newer.
- The Figma Desktop app.
- An MCP-compatible AI client such as Claude Desktop, Claude Code, Cursor or Codex.
Use it with UiChemy
When the Figma file is ready, UiChemy's Figma to WordPress turns it into an editable WordPress page. Get Started at uichemy.com.
Connect figma-pilot MCP: Run Figma API Code from an AI Agent to your AI client
claude mcp add figma-pilot -- npx @youware-labs/figma-pilot-mcpCheck prerequisites
Install Node.js 18 or newer and the Figma Desktop app.
Add the server
Run the claude mcp add command, or paste the JSON config into your MCP client's settings.
Connect Figma
Start the plugin in your Figma file, then ask the agent to run figma_status.
Client configuration
{
"mcpServers": {
"figma-pilot": {
"command": "npx",
"args": ["@youware-labs/figma-pilot-mcp"]
}
}
}What you can do with it
- Three-tool surfaceA small tool list keeps context free for the actual design work.
- Code executionThe agent writes JavaScript against the full Figma API.
- Batch editsModify many elements in one call instead of one tool call each.
- Filtered resultsTrim data in code before it reaches the model.
- Complex workflowsUse loops, conditionals and error handling in a single run.
Tools it exposes
figma_status
Checks the connection between the server and Figma.
figma_execute
Runs JavaScript with full access to the Figma API.
figma_get_api_docs
Returns the API documentation the agent needs to write correct code.
Frequently asked questions
What is figma-pilot?
figma-pilot is an open-source MCP server from Youware Labs that lets an AI coding agent control Figma by executing JavaScript. It exposes three tools only, status, execute and API docs, so the agent writes code instead of choosing from a long list of narrow tools.
How do I install figma-pilot?
Run claude mcp add figma-pilot -- npx @youware-labs/figma-pilot-mcp in Claude Code. For other clients, add a server named figma-pilot with the command npx and the argument @youware-labs/figma-pilot-mcp to your MCP config file, then connect it to Figma Desktop. Setup takes about a minute on most machines.
What do I need before using figma-pilot?
You need Node.js 18 or newer, the Figma Desktop app, and an MCP-compatible AI client such as Claude Desktop, Claude Code, Cursor or Codex. The plugin runs inside Figma Desktop, so the browser version of Figma is not enough. Open the file you want the agent to edit first.
Why does figma-pilot use code execution?
Each MCP tool definition takes up context window space. Three tools plus code execution means far fewer tokens spent on definitions, and the agent can batch changes, filter data and handle errors in one call. The README reports a saving of more than 90 percent in tool-definition tokens.
What can I build with figma-pilot?
The README shows two demos: an OpenAI-style landing page created from a prompt and a design system's components generated from a screenshot. In practice it suits any task the Figma API can do, from creating frames to bulk-updating styles. Check the README for the current tested client list.
Is figma-pilot free?
Yes. The repository uses the MIT licence and runs locally through npx. There is no hosted service. You will still need Figma Desktop and an AI client, and those follow their own plans and limits. Use figma_get_api_docs when the agent needs a reference.
How is figma-pilot different from other Figma MCP servers?
Most Figma MCP servers expose many specific tools, such as one to read a node and another to set a fill. figma-pilot exposes one execute tool and lets the agent write the logic. That trades a larger single call for a much smaller tool list. Compare it with Figma Unified MCP if you want more tools.


