Figma-Context-MCP server in n8n design data in automated workflows.

Launch the npm package without --stdio and it serves MCP over HTTP on port 3333. Run it somewhere your n8n instance can reach — the same host for a quick test, or a small always-on service next to self-hosted n8n. Get the token from Figma Settings > Security > Personal access tokens first:

In your workflow, attach an MCP Client Tool node to the Tools slot of an AI Agent node — it functions only there, not as a standalone step. Set the node's endpoint to your server's address (http://localhost:3333/sse for the SSE transport if both run on one host) and pick the matching transport in the node's settings. No credentials are needed at the node level; the Figma token was supplied when you launched the server.

Once connected, the node enumerates the server's tools — fetching translated file data and downloading image assets — and you can restrict which ones the agent may call.

A solid first test: a manual-trigger workflow with an AI Agent node, your MCP Client Tool attached, and a prompt that includes a Figma selection link (right-click a frame in Figma > Copy link to selection) asking the agent to list the frame's components and text content. The execution log should show a tool call out to the server with structured design data coming back.

Scope every request to a frame via selection links. Whole-file fetches on production design files produce payloads that eat the agent's context window and inflate token costs on every scheduled run.

There's no JSON file to break here, so failures are network and topology. If the node can't connect, check reachability from n8n's perspective: when n8n runs in Docker, localhost is the container, not your machine — use host.docker.internal or the host's LAN address in the endpoint URL. A connected node with an empty tool list usually means the transport selection doesn't match the endpoint path (SSE endpoint with Streamable HTTP selected, or vice versa).

If tool calls fail with authorization errors, the token you passed at server launch is wrong or lacks access to the linked file — fix it where the server runs and restart that process; nothing in n8n needs to change. Node-level errors land in the workflow execution log, on the node itself.

Figma-Context-MCP's live Loomal listing at https://loomal.ai/marketplace/figma-context-mcp shows its probed tool list — useful for planning which tools your workflow will allow. It tops the design tools category; for workflows that act on the web rather than on designs, the search and browser categories carry the natural companions.