Talk to Figma MCP
Author: grab
Description: Model Context Protocol (MCP) server + Figma plugin + WebSocket bridge that lets AI agents (Cursor, Claude Code) read Figma designs and modify them programmatically (annotations, layout, styling, components, text, connectors, export, etc.).
Stars: 6.5k
Forks: 698
License: MIT License
Category: Open Source
Overview
Installation
curl -fsSL https://bun.sh/install | bashbun setupbun socket{
"mcpServers": {
"TalkToFigma": {
"command": "bunx",
"args": ["cursor-talk-to-figma-mcp@latest"]
}
}
}powershell -c "irm bun.sh/install.ps1|iex"// uncomment this to allow connections in windows wsl
hostname: "0.0.0.0",bun socketFAQs
What are the 23 core tools available in Talk to Figma MCP and what does each one do?
The article mentions that the Talk to Figma MCP README documents 23 core tools but does not enumerate them. The referenced tools include get_instance_overrides, set_instance_overrides, set_annotation, set_multiple_annotations, and join_channel, plus bulk text replacement functionality. For the complete list and individual tool descriptions, consult the official GitHub README at the Grab repository.
How does Talk to Figma MCP's WebSocket bridge work and why is it required for write access?
The WebSocket bridge on port 3055 enables bidirectional communication between the external MCP server and the Figma plugin sandbox. Write operations require this architecture because Figma's plugin API only executes inside the Figma desktop application. The bridge relays agent commands to the plugin, which uses Figma's APIs to modify documents and returns results.
What are the most common issues faced when setting up an MCP server?
Common MCP server setup issues include environment PATH misconfigurations causing "command not found" errors, missing dependencies like absent Python modules or version conflicts, platform incompatibilities from OS differences, deployment failures from incomplete Docker images, and network problems such as port mismatches or health check failures. JSON syntax errors and invalid log levels also break startup.
What are the best practices for securing an MCP server?
Secure MCP servers by implementing OAuth 2.1 with RBAC, enforcing mTLS through reverse proxies, and isolating tools in sandboxed environments like gVisor or Firecracker VMs. Enable comprehensive logging, run weekly vulnerability scans using Trivy or Snyk, and conduct quarterly penetration tests targeting AI-specific attacks like prompt injection. Rotate credentials every ninety days.
How do I fix the 'empty results' error when using Talk to Figma MCP with my AI coding agent?
Start the WebSocket bridge before the MCP server by running `bun socket` first, then execute the `join_channel` command with your plugin's displayed channel name (for example, `join_channel with channel: '959ykpvn'`). If results remain empty, try sharing a Figma Dev link instead of a selection link, as some client configurations resolve Dev links more reliably.
What is the difference between Talk to Figma MCP and Figma Context MCP for design-to-code workflows?
Talk to Figma MCP enables bidirectional workflows where agents both extract and modify live designs, while Figma Context MCP optimizes token efficiency during one-way extraction. Choose Talk to Figma when your workflow includes programmatic design changes like bulk updates or component propagation. Choose Context MCP when you only need to read designs into code with minimal token overhead.