M
MCP Directory CURATED DIRECTORY
911 MCP servers indexed
中文
← Back to directory
T

Talk to Figma

Community
Let AI agents read, analyze, and modify Figma designs
How to install ↓ Compare GitHub source repository ↗
Category
Collaboration #29 of 55
Stars
★ 668 Popular
Transport
stdio (local process)
Runtime
Node.js · Bun · Docker
Credentials
No credential needed
License
MIT
Last commit
Tools
19
53FMRS · D

This project provides a feature-rich MCP server that enables deep integration between AI agents and Figma. It supports design analysis, element creation, and modification, and is compatible with multiple AI tools. The biggest advantage is that it works without a Dev Mode license. It is a practical tool for design teams and developers, but attention should be paid to multi-agent modification risks and the development status of some features.

Strongest · Documentation 13/20 Weakest · Reliability 8/20

Reliability
8/20
Security and permissions
10/20
Maintenance
11/20
Documentation
13/20
Setup experience
11/20
Why each score
Reliability 8/20
The README describes a complete WebSocket + Figma plugin + client architecture; the happy path (npx start, plugin import, channel-ID connection) looks plausible and a production-stable status is claimed. However, the supplied material contains no committed tests or CI workflows verifying key paths, which caps the score under the static-review rule; 9 issues are open; the multi-component chain (socket, plugin, channel) has many failure points; a troubleshooting doc exists but its contents are unverifiable. Score is 8, below the cap of 12.
Security and permissions 10/20
No real tokens or secrets appear in the install examples; no evidence of malware, credential theft, or covert exfiltration; connections are gated by a channel ID, and reading/modifying the Figma document is the openly stated purpose, so no red line is triggered. Deductions: no documented confirmation mechanism for destructive operations, least-privilege and plugin permission scope are not described, and the data flow (design content via local plugin and WebSocket to the AI client) is only partially disclosed, especially with multi-agent parallel modification, hence 10.
Maintenance 11/20
Clear MIT license, not archived; a CHANGELOG, contributing guide, and releases with .dxt assets indicate a release cadence; the README lists active development items (Figma Variables, Tailwind/SwiftUI export). Deductions: no visible security-response channel, commit frequency and issue-response pace cannot be verified (9 open issues), and no dependency-update evidence is in scope. Stars count only as a discovery signal and add no points.
Documentation 13/20
Documentation is layered: README plus INSTALLATION, COMMANDS, TROUBLESHOOTING, CONTRIBUTING, and CHANGELOG; client-specific config examples are given (Claude Desktop .dxt, Cursor JSON) and limitations are disclosed (set_current_page blocked, explicit parentId required) along with multi-agent behavior. Deductions: no explicit cost/quota or limits section, troubleshooting content is unverifiable, and there are hidden assumptions (Figma Desktop required, manual channel-ID flow).
Setup experience 11/20
An npx all-in-one command, plugin import steps, and concrete client config examples for Claude Desktop (.dxt auto-config) and Cursor (JSON) are provided, with a Docker alternative. Deductions: the overall flow remains multi-component and manual — Figma Desktop and plugin import are required, and every session requires copying a channel ID and having the agent connect manually; the README mixes bun/npm run commands, adding hidden prerequisites; with no CI/test execution evidence, the static cap of 15 applies, so the score is 11.

Static review · not runListed 2026-08-07

Read the FMRS scoring method →

Fit and risk

What it can accessUses the network

Best for

  • UX/UI teams: Automate repetitive design tasks and maintain brand consistency
  • Developers: Generate production-ready code directly from designs
  • Teams that need a free Figma account alternative

Not for

  • Users needing full support for Figma Variables (still in development)
  • Users needing advanced export to Tailwind CSS/SwiftUI (coming soon)

Required permissions

  • Requires Node.js and Figma Desktop
  • Requires network access to run npx and connect WebSocket
  • Requires installing the Figma plugin and granting it access to the current design file
  • AI tool must be configured to connect to the MCP server

Risks and side effects

  • Multiple agents modifying the document simultaneously may cause conflicts, though the server has a command queue, it's recommended to explicitly specify the parentId parameter
  • There may be undiscovered bugs as the project is under active development
  • Dependencies on third-party npm packages and Figma plugins; ensure sources are trusted

Setup

Before you start

Runtime：Node.js · Bun · Docker

  1. Run npx claude-talk-to-figma-mcp to start the WebSocket server.
  2. Install the plugin in Figma Desktop (Menu → Plugins → Development → Import plugin from manifest, select src/claude_mcp_plugin/manifest.json).
  3. Configure your AI tool:
  • Claude Desktop: Download the .dxt file and double-click to auto-configure.
  • Cursor: In settings, create a new MCP server with command npx -p claude-talk-to-figma-mcp@latest claude-talk-to-figma-mcp-server.
  • Other tools: Follow the detailed installation guide.
  1. Open the plugin in Figma, copy the channel ID, and tell your AI: "Connect to Figma, channel {your-ID}".

Check that it works

Import src/claude_mcp_plugin/manifest. as a development plugin in Figma Desktop, copy the channel ID, and type "Connect to Figma, channel {your-ID}" in your AI client. Then ask the client to run get_document_info; returning document details confirms the connection works.

Troubleshooting

  1. Ensure Node.js is installed and the npx command runs correctly
  2. Check that the Figma plugin is correctly installed and opened
  3. Ensure the channel ID is correctly copied and entered in the chat
  4. Refer to TROUBLESHOOTING.md for common error solutions

Things to try

Once connected, you can ask your AI assistant things like:

  • Get the current Figma document info and my current selection
  • Scan all text with contrast ratio below 4.5:1 and suggest colors that meet WCAG AA
  • Change #FF6B6B to #E63946 in all primary buttons throughout the document
  • Generate a React component for CardProduct with styles and export its assets

Tools 19

get_document_info read-only
Get basic information about the Figma document
get_current_selection read-only
Get the currently selected elements
get_styles read-only
Get styles defined in the document
scan_text read-only
Scan text elements in the document
audit_components read-only
Audit component usage in the document
export_assets read-only
Export design assets
create_shape writes
Create a shape element
create_text writes
Create a text element
Show 11 more tools
create_frame writes
Create a frame element
clone_element writes
Clone an existing element
group_elements writes
Group multiple elements
modify_colors writes
Modify element colors
modify_borders writes
Modify element borders
modify_corners writes
Modify element corner radius
modify_shadows writes
Modify element shadows
auto_layout writes
Apply auto layout
advanced_typography writes
Set advanced typography
set_local_component writes
Convert element to local component
set_team_component writes
Use team library component

Use cases

Automate accessibility audits, detect contrast issues and suggest fixes
Bulk update styles like colors, typography, and spacing
Analyze visual hierarchy and suggest improvements
Generate React/Vue/SwiftUI components from designs
Keep design and development in sync, reduce handoff friction

Supported clients

Claude Desktop
Claude Code
Cursor
Windsurf
VS Code
GitHub Copilot
Antigravity
Cline
Roo Code

Listed from the project's documentation, not tested by this site.

Overview

This is a Model Context Protocol (MCP) server that enables AI tools like Claude Desktop, Claude Code, Cursor, Antigravity, etc. to interact with Figma design files. It supports design analysis, element creation, style modification, and works with any Figma account (even free ones) without needing a Dev Mode license. The server communicates with a Figma plugin via WebSocket and supports safe parallel execution for multiple AI agents.

Similar servers

Compare with other MCP servers

Source revision c7adf243fc29 Data synced 2026-10-12 Read the FMRS scoring method