Brilliant MCP
OfficialThis server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Brilliant MCP
Your coding agent designs with you on a real vector canvas. This is the local MCP server for Brilliant, the design tool whose canvas you and your agent both edit: the agent lays out the screen, you drag, retype and restyle it, and the agent sees your changes and keeps going.
@brilliant-hq/mcp is a small local router. It connects any MCP client over stdio
to a Brilliant tab in your browser or to the Brilliant desktop app, entirely on
127.0.0.1. Zero bytes of MCP or design data touch Brilliant's servers.
Install
Client | One line |
Claude Code |
|
Cursor |
|
VS Code |
|
Codex |
|
Gemini CLI |
|
Anything else |
|
install <client> knows claude-code, cursor, vscode, codex, gemini, antigravity,
windsurf, zed, opencode, amp and claude-desktop. Any other MCP client takes the
standard stdio entry:
{ "mcpServers": { "brilliant": { "command": "npx", "args": ["-y", "@brilliant-hq/mcp"] } } }Requires Node 20 or newer. No API key.
Related MCP server: Canvas AIGC
First run
Ask your agent for a design: "design a pricing page for this app". With no
Brilliant open, the first tool call opens brilliant.design in your browser
(Chrome, Edge or Firefox) and pairs it with your agent session; allow the one-time
local network prompt. If the Brilliant desktop app is running, the server connects
to it instead and opens nothing. Listing the tools never opens anything.
What your agent can do
The tools are Brilliant's own, served by the tab or the app; this package relays them unchanged. The main doors:
init: start a session and receive Brilliant's documentation for its tools.edit: write designs, as HTML with inline CSS or as Blueprint, Brilliant's plain-text design format. Everything lands as editable vector layers.read,get_selection: read canvases, design systems and what you selected.export: PNG, JPEG, WebP, SVG, PDF, HTML and React.get_knowledge: load design and format guidance on demand.generate_image,generate_svg,vectorize_image,remove_background: image and vector generation inside a design.
Every tool carries MCP annotations (a title, read-only and destructive hints), so your client can tell a read from a write before it calls.
How it works
The server is a relay: MCP JSON-RPC crosses it verbatim. MCP semantics live in the Brilliant tab or desktop app, never here.
The first instance on a machine becomes the leader (one tab socket, one permission grant); every other MCP client's instance is a follower that forwards over loopback. Leadership reassigns when the leader exits.
The Brilliant desktop app is detected on
127.0.0.1:3333to3335and covered by the same entry.
The full wire contract is in PROTOCOL.md.
Privacy
This package makes no network calls of its own: no telemetry, no analytics, no
update checks. It talks only to a brilliant.design tab you approve or to the
Brilliant desktop app on your machine, over 127.0.0.1.
One thing does reach Brilliant's website: when the server opens the pairing tab,
the link carries three attribution params, utm_source=mcp-helper, your MCP
client's name (utm_medium, for example claude-code) and, if the install
snippet set one, a door tag (utm_content, from --src <tag> or
BRILLIANT_MCP_SRC). They tell us which install path people use. They say
nothing about you, your machine or your designs, and the pairing token stays in
the URL fragment, which a browser never sends to a server.
It keeps a small state directory at ~/.config/brilliant/mcp-helper (its ports, the pairings
you approved and the last tool list), readable only by you.
Brilliant itself, like any app, has its own data handling: the image tools run on Brilliant's servers when your agent calls them, and a cloud project syncs when you choose one. See the privacy policy.
Security
Every listener binds the literal
127.0.0.1: never0.0.0.0, never a scan of private ranges.The tab socket validates
Origin: https://brilliant.design, a loopbackHost(against DNS rebinding), and a pairing token or an association you approved.The pairing token is never served from any local endpoint; the state files are mode
0600, and the web side never reads them.
To report a vulnerability, see SECURITY.md.
CLI
npx @brilliant-hq/mcp # run the stdio MCP server (default)
npx @brilliant-hq/mcp --src <tag> # run it, tagging the pairing link with an install door
npx @brilliant-hq/mcp --version
npx @brilliant-hq/mcp install <client> # add the server to <client>'s MCP config
npx @brilliant-hq/mcp install <client> --dry-run # preview the change, write nothing
npx @brilliant-hq/mcp install <client> --remove # remove it againinstall merges into the client's config without touching other servers, keeps
JSONC comments, and --remove is its exact inverse.
Development
npm install
npm testPricing
Canvas work by your own agents over MCP is unlimited on every Brilliant plan, the free plan included. Generated images and background removal draw on your Brilliant AI usage. Details: https://brilliant.design/pricing
Licence
The code in this repository is licensed under Apache-2.0. The licence
covers this local server only: the Brilliant editor, engine and services are not
open source, and the Brilliant name and logo are not licensed (see NOTICE).
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Generate images from Claude, ChatGPT, Cursor and Codex with 12 top AI image models and 10 tools: up to 4 images per call, SVG vectorization, editing and more. Remote MCP with OAuth, no API keys. Pay per image, no subscription.
Generate images from Claude, ChatGPT, Cursor and Codex with 12 top AI image models. Up to 4 images per call, SVG vectorization, background removal, upscaling and image-to-video. Remote MCP with OAuth, no API keys.
The Canva MCP server connects AI assistants (like Claude, ChatGPT, and Cursor) to Canva's API, enabling them to create and manage designs directly within chat conversations. Key capabilities include generating new designs from prompts, autofilling templates, searching and resizing existing designs, importing files from URLs, exporting designs as PDFs or images, and managing folders and comments without switching between tools.
Connect AI agents to Flato's editable canvas runtime through a hosted MCP server.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceMCP server that enables AI coding agents to read and write to a local-first HTML/CSS design canvas, bridging visual design and code generation.1MIT
- AlicenseNot gradedqualityDmaintenanceA MCP server that bridges web canvas with local Codex/Claude, exposing 23 tools for reading, creating, modifying, connecting, and generating canvas content via local HTTP bridge and stdio MCP.AGPL 3.0
- AlicenseNot gradedqualityAmaintenanceEnables AI agents to read and write Figma designs through a local WebSocket relay, turning Figma selections into framework-aware code and building/editing designs directly on the canvas. Provides 112 MCP tools for bidirectional design-code workflows with support for any MCP client.MIT
- AlicenseNot gradedqualityAmaintenanceOpen-source multiplayer design canvas (alternative to Paper.design) with a built-in MCP server. Agents create and stream HTML artboards live next to human collaborators, review them via screenshots, comment on elements, search icons/images/logos and follow shared design guidelines.74 npm1,909AGPL 3.0