mcp server
figma-mcp-server-actor
An MCP that enables AI assistants to interact programmatically with Figma designs
Set up
Runs on your computer. Pick your agent and operating system.
Before you start
- Node.js 20 or newer
{
"mcpServers": {
"figma-mcp-server-actor": {
"command": "npx",
"args": [
"-y",
"@oabolade-pegasus/figma-mcp-server"
],
"env": {
"FIGMA_ACCESS_TOKEN": "<your FIGMA_ACCESS_TOKEN>"
}
}
}
}same format Cursor, GitHub Copilot CLI, Junie, JetBrains AI Assistant, Augment (Auggie CLI), Factory Droid, Kiro, Devin CLI +12
Values in angle brackets are placeholders. Restart Claude Desktop after editing its config.Claude Desktop docs
README
synced Oct 14, 2026Figma 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:
| Input | Purpose |
|---|---|
oauthClientId | From Figma → Developers → Apps |
oauthClientSecret | Same app |
oauthRedirectUri | Exact redirect URL registered on the app, e.g. https://<your-run>.runs.apify.net/oauth/callback |
oauthScopes | Optional; comma-separated scopes. Must be a subset of the app’s configured scopes. Defaults include file_content:read, file_content:write, file_comments:read, file_comments:write. |
Flow
- Start the Actor with OAuth fields filled and no PAT.
- Open
GET https://<container-url>/oauth/authorizein a browser (normal browser — not an embedded WebView; Figma requirement). - 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)
| Parameter | Type | Default | Description |
|---|---|---|---|
figmaAccessToken | string | — | PAT (recommended). If set, used for all Figma API calls. |
port | integer | 8080 | HTTP listen port (Apify may override via web_server_port). |
oauthClientId | string | — | OAuth app client ID (only if not using PAT). |
oauthClientSecret | string | — | OAuth app secret (secret input). |
oauthRedirectUri | string | — | Must match Figma app redirect URI exactly. |
oauthScopes | string | — | Optional scope override (comma-separated). |
maxConcurrentRequests | integer | 10 | Reserved / future use. |
enableCaching | boolean | true | Cache GET responses from Figma. |
Local / CLI input file example (apify run):
{
"figmaAccessToken": "figd_...",
"port": 8080
}
HTTP API
| Method | Path | Description |
|---|---|---|
GET | / | Service info and endpoint list |
GET | /health | Liveness; includes auth_mode: pat | oauth | pending_oauth, figma_authenticated |
POST | /mcp | MCP JSON-RPC 2.0 (primary transport for this server) |
GET | /oauth/authorize | Start OAuth (browser); disabled if PAT is set |
GET | /oauth/callback | OAuth redirect handler |
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
- Deploy and start the Actor; copy the run Container URL (while status is RUNNING).
- 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_TOKENinmcp.jsonis not passed to this Node process; it does not replace Figma auth. UsefigmaAccessToken(or OAuth) in Actor input.- Container URLs are per run and stop working when the run ends; update
mcp.jsonand 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/authorizeand/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,
/mcpJSON-RPC.
References: Apify Actors · Figma REST API · Model Context Protocol