Figma MCP Server (Apify Actor)

Model Context Protocol (MCP) server that exposes Figma REST API capabilities to AI assistants (Cursor, Claude, custom agents). Runs as a long-running web server on the Apify platform (standby / container URL) or locally via apify run.

Features

• File analysis — structure, metadata, styles, depth-limited trees
• Components — list, details, usage
• Assets — export nodes (PNG, SVG, PDF, JPG)
• Comments — read and create
• Projects / teams — listings and metadata
• MCP resources & prompts for guided workflows

Authentication

Personal Access Token (default)

Recommended for solo use. Set figmaAccessToken in Actor input (or FIGMA_ACCESS_TOKEN in env).

• Generate: Figma → Settings → Security → Personal access tokens
• API header: X-Figma-Token
• When PAT is set, OAuth is disabled for API calls (and /oauth/* routes return an error).

OAuth 2.0 (optional — team / shared files)

Use when you omit figmaAccessToken and provide a Figma OAuth app:

Flow

1. Start the Actor with OAuth fields filled and no PAT.
2. Open GET https://<container-url>/oauth/authorize in a browser (normal browser — not an embedded WebView; Figma requirement).
3. After consent, Figma redirects to /oauth/callback; tokens are stored in the run’s default key-value store (FIGMA_OAUTH_SESSION) and refreshed when near expiry.

Token exchange uses Figma’s documented endpoints (https://api.figma.com/v1/oauth/token and .../oauth/refresh). Codes expire quickly — complete the redirect promptly.

Requirements

• Node.js ≥ 20
• Apify account (for cloud)
• Figma PAT and/or OAuth app

Quick start (Apify Cloud)

npm install
apify login
apify push

In Apify Console, open the Actor, set input (at minimum figmaAccessToken), start with web server / long-running as required by your template, then use the Container URL from the run.

Deploy reference: apify push builds and publishes; see Apify CLI.

Configuration (input)

Local / CLI input file example (apify run):

{
  "figmaAccessToken": "figd_...",
  "port": 8080
}

HTTP API

Apify MCP path: .actor/actor.json sets webServerMcpPath to /mcp. Your MCP client URL is typically:

https://<container-id>.runs.apify.net/mcp

(Use the Container URL from the run page + /mcp.)

Cursor IDE

1. Deploy and start the Actor; copy the run Container URL (while status is RUNNING).
2. Edit ~/.cursor/mcp.json (paths differ on Windows/Linux — see Cursor docs).

Example (remote HTTP MCP — adjust to your Cursor version’s expected shape):

{
  "mcpServers": {
    "figma-mcp-standby": {
      "url": "https://YOUR_RUN_ID.runs.apify.net/mcp"
    }
  }
}

Notes

• env.APIFY_TOKEN in mcp.json is not passed to this Node process; it does not replace Figma auth. Use figmaAccessToken (or OAuth) in Actor input.
• Container URLs are per run and stop working when the run ends; update mcp.json and restart Cursor after a new run.
• If connection fails, use Cursor → Output → MCP and curl https://.../health.

Claude Desktop / other clients

Use your client’s supported way to attach a remote MCP URL (e.g. mcp-remote pointing at https://.../mcp). Paths and query params depend on the client version.

Local development

npm install
apify run

Or run the entry with Node if you inject env / input yourself:

export FIGMA_ACCESS_TOKEN="figd_..."
node src/main.js

Check health (port from logs or input):

curl -s http://localhost:8080/health | jq .

Sample MCP initialize:

curl -s -X POST http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{}}'

Project structure

figma_mcp_actor/
├── .actor/
│   ├── actor.json           # Apify Actor metadata (incl. webServerMcpPath: /mcp)
│   ├── input_schema.json
│   ├── output_schema.json
│   └── dataset_schema.json
├── src/
│   ├── main.js              # Express app, /mcp, /oauth/*, Apify bootstrap
│   ├── figma/
│   │   ├── auth.js          # PAT + OAuth token exchange, refresh, KV hook
│   │   └── client.js        # Figma REST wrapper
│   ├── oauth/
│   │   └── persist.js       # Apify KV: FIGMA_OAUTH_SESSION
│   ├── mcp/                 # Protocol, handlers, server
│   ├── tools/
│   ├── resources/
│   └── prompts/
├── Dockerfile
├── package.json
└── README.md

MCP tools (overview)

Includes (names may vary slightly in code): file analysis (analyze_file, get_file_structure, …), components (list_components, …), export (export_node, …), comments (get_comments, create_comment, …), design modification helpers, plus resources (figma://file/..., etc.) and prompts for common tasks.

Limitations

• REST vs Plugin API — deep design writes may need the Plugin API; this Actor uses REST where supported.
• Ephemeral Apify URLs — each run has a new base URL unless you use a stable hosting pattern.
• OAuth — single stored session per run (KV); not multi-user concurrent OAuth in one process.

Scripts

npm run lint
npm run lint:fix
npm run format

License

ISC

Changelog

0.2

• Optional OAuth 2.0 (Figma authorization code + refresh); PAT remains default when set.
• /oauth/authorize and /oauth/callback; session persisted in default KV (FIGMA_OAUTH_SESSION).
• Input: oauthRedirectUri, oauthScopes.
• Health JSON: auth_mode, figma_authenticated.

0.1 / 0.0.1

• Initial MCP + Figma REST integration, PAT, long-running server, /mcp JSON-RPC.

References: Apify Actors · Figma REST API · Model Context Protocol