Figma Context MCP
Give your coding agent access to your Figma data. Implement designs in any framework in one-shot.
Description from the maker (original English)
The Figma file you have open, as MCP tools. Extract, edit, and keep a saved set.
These three are never merged into one score: opinions, measurements and popularity are different things.
We have not written a plain-language summary for this one yet. Below are the facts we read from the registry.
Pick the AI app you use, then copy the settings.
FIGSNAP_MCP_URL, FIGSNAP_MCP_TOKENValues marked secret are keys to your account. Never share them or paste them into a chat.
You need
Which app do you use?
Pick your app above. We remember it, so next time the button says “Install in …”.
Open Claude Desktop → Settings → Developer → Edit Config, and paste this into claude_desktop_config.json
{
"mcpServers": {
"figsnap": {
"command": "npx",
"args": [
"-y",
"figsnap-mcp"
],
"env": {
"FIGSNAP_MCP_URL": "your-value",
"FIGSNAP_MCP_TOKEN": "your-value"
}
}
}
}Run this in your terminal
claude mcp add figsnap --env FIGSNAP_MCP_URL=your-value --env FIGSNAP_MCP_TOKEN=your-value -- npx -y figsnap-mcpFill in the empty values before running it.
Paste into .cursor/mcp.json in your project, or ~/.cursor/mcp.json
{
"mcpServers": {
"figsnap": {
"command": "npx",
"args": [
"-y",
"figsnap-mcp"
],
"env": {
"FIGSNAP_MCP_URL": "your-value",
"FIGSNAP_MCP_TOKEN": "your-value"
}
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
Paste into .vscode/mcp.json in your project
{
"servers": {
"figsnap": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"figsnap-mcp"
],
"env": {
"FIGSNAP_MCP_URL": "your-value",
"FIGSNAP_MCP_TOKEN": "your-value"
}
}
}
}Your browser asks to open the app, then the app shows the settings and asks you to confirm. Key values are left empty for you to fill in.
This one runs on your computer, and ChatGPT can only connect to servers on the internet. Pick another app.
Your choice stays on this device in a cookie; no account needed.
FIGSNAP_MCP_URL envThese settings are generated from the maker’s registry entry. If the README says otherwise, follow the README.
Runs on your computer; the tool list appears in your AI app after you install it.
We check information anyone can see; we have not read the code. Checks that do not apply are left out of the count. · Checked Nov 8, 2026
The registry name was verified with the GitHub account arimunandar, which owns the repo
We have not read this repo from GitHub yet
We have not read this repo from GitHub yet
This one runs on your computer; there is no internet endpoint
Runs on your computer and uses local settings instead
Runs on your computer; we can only read tool lists from internet endpoints
Compared with every active server in this category in the registry. Stars count only when the publisher owns the repo.
It passed 1 of 1 applicable basic checks. We have not read the code. Before adding a key or token, see what access the server asks for and use a key with only the permissions you need.
You need to fill in: FIGSNAP_MCP_TOKEN
It runs on your computer. Your AI app starts it when needed, and data does not pass through the maker’s servers (unless it calls an online service itself).
The GitHub repo last changed –, and the latest version (0.2.0) reached the registry 2 months ago.
Apps that support MCP, such as Claude Desktop, Claude Code, Cursor and VS Code. You also need Node.js on your computer.
In the same category: Figma Context MCP, Claude Real Video, Figma MCP Server
Servers in the same category from other makers, sorted by stars.
Reading a card:★ GitHub starsiThree pillars: reviews · checks · usageiUpdated = latest version published to the registryi
Give your coding agent access to your Figma data. Implement designs in any framework in one-shot.
Let any LLM watch a video locally — and search everything it has ever watched.
Figma’s official server gives AI your design data so it can turn designs into code or describe them accurately. Connects to Figma’s cloud.
Copied from the maker’s repo as written. PopNx MCP has not reviewed or endorsed it. HTML, images and scripts are removed. Links lead off this site.
The Figma file you have open, as MCP tools — and nothing else.
claude mcp add figsnap-mcp -s user -- npx -y figsnap-mcp
Those install the MCP server. It answers nothing until the daemon is running and the Figma plugin is open — see Getting it running, which is five steps and takes about two minutes.
A Figma plugin, a local daemon, and an MCP server. Your coding agent reads the design, extracts a node as PNG, React, HTML or CSS, edits the canvas when you allow it, and shares a saved set of components with you. No relay, no account, no network beyond loopback.
MCP client figsnap-mcp daemon the plugin
(Claude Code) stdio (stateless HTTP :3058 ws (open in Figma) ─────────────────► proxy) ──────────► /panel ──────────► figma.* 127.0.0.1 only
Three processes and one rule: figma.* exists only while the plugin is open, so the plugin holds the connection and everything else asks questions down it.
This repo is a derivative of Figsnap, cut down to one job. Figsnap ships four products from one package.json — the plugin, a Cloudflare Worker relay with accounts, an ACP chat client, and this MCP server. If all you want is MCP tools and a saved set, you had to clone all of it, build a 4,200-line panel you would never open, and load a plugin whose manifest whitelists a Worker you would never sign in to.
So: no ACP client, no harness discovery, no session store, no relay, no accounts, no chat, no API browser, no code-export UI. What is left is the extraction engine, the 41 tools, and a panel with three panes.
Figsnap is not modified by any of this, and the two can run side by side — this daemon is on port 3058 with its token in ~/.figsnap-mcp/agent-token, where Figsnap's is on 3056 with its token in ~/.figsnap/agent-token.
One caution on that manifest: Figma validates allowedDomains and refuses an IP literal — Invalid value for allowedDomains. 'ws://127.0.0.1:3058' must be a valid URL — and a manifest it refuses is a manifest it does not load, so the plugin quietly keeps running the last good one. Name localhost, and check the console after any change here, because nothing else reports it.
They are kept apart in Figma too. clientStorage is keyed by plugin id, so the manifest carries an id of its own rather than Figsnap's REPLACE_ON_PUBLISH, and every key this plugin writes is prefixed figsnap-mcp: on top of that. Sharing either one means sharing the stored daemon address, and a panel that inherits ws://localhost:3056 is then blocked by its own manifest's CSP — with a console error the designer never sees. The panel also refuses any stored or typed address outside the port its manifest allows, and drops the token that came with it.
npm install -g figsnap-mcp
Two commands come with it:
figsnap-mcp-daemon | the bridge; leave it running while you work |
figsnap-mcp | the MCP server, spawned by your client — not run by hand |
You can skip the install and let npx fetch it, which is what an MCP client config usually does. The Figma plugin ships in the package too: after a global install its manifest is at
npm root -g # …/lib/node_modules — the manifest is figsnap-mcp/manifest.json
Or clone the repository and npm install && npm run build, which is the same thing with the sources beside it.
Five steps, once.
1. Build it.
npm install -g figsnap-mcp
From a clone instead: npm install && npm run build.
2. Start the daemon. Leave it running; it is the only thing that talks to Figma.
figsnap-mcp-daemon # --allow-edits opens the writing tools at boot
From a clone: npm run daemon. figsnap-mcp-daemon --help lists the rest.
It prints its address and its token:
figsnap-mcp-daemon 0.1.0
panel socket ws://127.0.0.1:3058/panel http http://127.0.0.1:3058 token 8fT2qN4vRk1pXwLzYc7BhJ0mAeUdSg9T edits off — turn them on in the plugin, or start with --allow-edits
3. Load the plugin. In Figma desktop: Plugins → Development → Import plugin from manifest, pick the package's manifest.json (npm root -g finds it, or it is in the repository root), then run Figsnap MCP.
4. Pair the panel. Copy the token line from step 2 into the panel's Connect pane and press Connect. The dot turns green and the Address reads ws://localhost:3058/panel.
This is the only place a token is ever typed. It is stored in Figma's own per-user storage, so the panel reconnects itself every time from now on.
5. Register the MCP server.
claude mcp add figsnap-mcp -s user -- npx -y figsnap-mcp
Restart the client, then claude mcp list shows figsnap-mcp · ✔ Connected. From a clone, name the file instead: -- node /path/to/FigsnapMCP/agent/mcp-stdio.mjs.
Check the whole chain:
curl -s http://127.0.0.1:3058/health
panelConnected: true is the line that matters — it means Figma is on the other end.
An MCP client never needs it. This is the part that surprises people: claude mcp add takes no token, no environment variable, no config. The MCP server reads the daemon's own file on the way past.
Only two things use the token, and only one of them is you:
| Who | How it gets it |
|---|---|
| The Figma panel | You paste it, once, in Connect |
agent/mcp-stdio.mjs | Reads ~/.figsnap-mcp/agent-token by itself |
Where it comes from. The first time the daemon starts it makes one — 24 random bytes, base64url — and writes it to ~/.figsnap-mcp/agent-token with mode 600. Every later start reads that same file back, so the token is stable and the panel is not re-paired every morning.
How to see it again without restarting anything:
cat ~/.figsnap-mcp/agent-token
Why there is one at all. The daemon listens on a loopback port, and any web page you happen to visit can open a socket to localhost. Two things stop it: the Origin header, checked on upgrade, which a browser cannot forge; and this token, because a browser WebSocket cannot set headers. Only GET /health is reachable without it — so the panel can tell you the daemon is running before it has been paired.
If it leaks, rotate it:
npm run daemon -- --new-token
That writes a fresh one and invalidates the old. Re-paste it in Connect; MCP clients pick the new one up on their own, because they read the file.
To use one of your own — a fixed token in a script, say — set FIGSNAP_MCP_TOKEN and the daemon uses it instead of the file. Set the same variable for the MCP client if it cannot read your home directory. FIGSNAP_MCP_URL moves the address the client dials.
claude mcp add figsnap-mcp -s user -- npx -y figsnap-mcp
For a client that takes JSON:
{ "mcpServers": { "figsnap-mcp": { "command": "npx", "args": ["-y", "figsnap-mcp"] } } }
figsnap-mcp-daemon --mcp prints both, with paths already filled in.
Then claude mcp list should show figsnap-mcp · ✔ Connected, and in a session figma_get_selection answers about whatever is selected on the canvas.
Three things can be wrong, and each says so differently:
| What the tool says | What to do |
|---|---|
No figsnap-mcp daemon at http://127.0.0.1:3058 | npm run daemon |
The Figsnap MCP plugin is not open in Figma | Open the file and run the plugin |
The figsnap-mcp daemon rejected the token | cat ~/.figsnap-mcp/agent-token, or set FIGSNAP_MCP_TOKEN |
Editing the file is switched off | Turn on Allow edits in the plugin's Tools pane |
Three things a question about a Figma file almost always needs are addressable rather than called for, so a client can @-mention them:
| URI | What it is |
|---|---|
figma://selection | Everything selected, extracted |
figma://page | The layer tree, three levels deep |
figma://library | Components, styles and variables, with ids |
figma://node/{nodeId} | One layer — figma://node/21:10314 |
Four panes, and the one it opens on is Selection.
Selection is a preview of whatever you have picked on the canvas — the picture, its name, type, size, child count and node id — with a folder picker and a Save button beside it. Saving from here is the short way round: no switching panes, no hunting for the layer again in a list.
The preview costs one PNG export. The panel used to run a full extraction on every selection change — HTML, TSX, two stylesheets and a 2× image — for a picture nobody was looking at; the code outputs are what MCP asks for, on request, not what a click costs. A small layer is magnified at most 4×, so an icon and a screen do not both fill the stage.
Minimise (the ▼ at the top right) drops the window to a 44px strip and the preview under it, so the canvas is clear while the plugin keeps running — the daemon needs the panel open, but you do not need to look at it. The strip still names what is selected and still has Save, because those are the two things worth having while you work. ▲ puts it back to whatever size you had dragged it to.
Connect and Tools are pairing and the tool list; Saved is the folder manager. Both are described below.
41 of them: 14 read Figma, 26 write to it, and one is answered in the daemon. The Tools pane lists them all with the writing ones marked; GET /tools is the same list as JSON.
Reading is always allowed. Writing is not, until you say so — every mutating tool is refused until *Allow edits* is on, which is a switch the designer holds rather than a prompt the agent can talk past. A harness running with permission prompts disabled still cannot get past it. --allow-edits opens the same gate from the terminal, for when the work is happening there.
Two tools fold thirteen plugin commands into one argument each, because 41 tool descriptions already cost real context on every request:
figma_extract takes nodeId, or nodeIds, urls, selection: true, saved: true for a batch.figma_saved takes an action: list, folders, save, unsave, clear, move, newFolder, renameFolder, deleteFolder.A picture comes back as a real image block, never as base64 in a text field — that is what figma_export_png is for, one node at a time.
figma_ios_spec answers with the properties Figma actually holds rather than with CSS, in points: per-axis sizing, auto layout and its padding, stroke weight and alignment, corner smoothing, effects, the text block — and every fill and stroke marked painted or not. Reach for it when the answer becomes code. `fig
Only the beginning is shown; read the rest in the repo