Figma Edit MCP

Connect AI assistants to Figma via Model Context Protocol to read designs, create and modify elements, and manage design systems programmatically.

This plugin empowers your AI assistant to become a Figma assistant, executing design updates Safer, Cleaner, and Faster than a human ever could.

This plugin allows you as a Designer to focus purely on creative decision-making, leaving the error-prone & repetitive manual changes to the automated systems.

📖 Documentation: neozhehan.github.io/figma-edit-mcp

Read more about our design philosophy here.

Core Principles

• 🛡️ Safer: The plugin performs programmatic checks and protections that exceed those in the Figma Desktop app. For example, it prevents the deletion of variables that are still in use, avoiding dangling references. By enforcing these strict validations before an action is taken, the plugin protects designs from both human error and AI hallucinations.
• ✨ Cleaner: Programmatic, thorough operations mean no node is ever skipped or forgotten during large updates, ensuring that the design file is always consistent.
• ⚡ Faster: Executing batch operations (like bulk text replacement or instance override propagation) via AI reduces hours of tedious manual design work down to seconds.

Safer than Figma Itself

Every other way to give an AI write access to your Figma file trusts the AI to behave. This one doesn't — and that's the whole point.

An AI agent is one hallucination away from deleting the wrong layer, overwriting a shared component, or quietly mangling a design-system token — and you may not notice until the damage is downstream. Figma Edit MCP treats the agent as untrusted with respect to edit safety, and puts a gate it cannot bypass in front of every write. The agent never decides what's safe; the plugin does, and it refuses anything outside the lines — with a structured error, not a guess.

What the agent physically cannot do to your file:

• 🎯 Wander off — it only writes inside the region you point it at; no scope, no edits.
• 👻 Edit a hallucinated layer — every write must name the exact node it's touching; a stale or invented ID is refused, never guessed.
• 💥 Half-finish a bulk edit — a batch with one bad target changes nothing; all-or-nothing, never a partial mess.
• 🔒 Touch what you protected — locked layers, shared-library assets, and the insides of component instances are off-limits.
• 🧨 Saw off the branch it sits on — it can't delete or replace the anchor of your editing session.
• 🌍 Reach for global tokens uninvited — editing your variables or styles takes a separate, explicit opt-in, off by default.

And it's honest about its limits: it guarantees where, which node, and what kind of edit happens — not whether a valid, in-bounds edit was the one you actually meant. You stay in charge of intent; it removes the accidents.

📖 The full contract — every guarantee, the conditions it holds under, and exactly how each is enforced — lives in SAFETY.md. (Your agent loads the same rules at runtime via the figma-edit skill or figma-edit://guide/*.)

Supported AI Integrations

• Cursor
• GitHub Copilot (VS Code)
• Google Antigravity
• Claude Code (VS Code & CLI)
• Claude Desktop (Chat, Cowork & Code)
• LM Studio

Quick

... View full README on GitHub