claude.design-mcp
Drives the real Claude Design web app from your editor or agent, enabling creation, iteration, and retrieval of designs generated on your own claude.ai account via automated browser interaction.
README
claude.design-mcp
An MCP that drives the real Claude Design web app from your editor/agent — log in once, then create, iterate on, and pull designs that claude.ai/design generates on your own account (not a local imitation).
How it works
- It drives your own logged-in Chrome (a dedicated profile) over CDP with
playwright-core, and talks to the realclaude.ai/design"Omelette" API as you, through your browser session. - Generation is triggered the way the website does it — your prompt is typed into the
design composer and submitted; the tool then waits for the turn to finish (the
ReleaseTurnnetwork signal + file-tree stability) and reports the files Claude Design wrote. Files are pulled back to local on request. - Project metadata, files, deletes, and direct file edits use the documented JSON RPCs
(
CreateProject/ListFiles/GetFile/WriteFiles/EditFile/DeleteProject), run in-page so they share your session + Cloudflare clearance. - Not a
claude -pmimic. Every design is produced by claude.ai/design itself.
Tools
| Tool | Does |
|---|---|
design_login |
One-time: open Chrome to log into claude.ai/design (session persists) |
design_list |
List your claude.ai/design projects |
design_create |
Create a project and generate a design from a prompt — prompt, name? |
design_iterate |
Send a follow-up prompt to modify a design — projectId, prompt |
design_pull |
Download a project's files to local — projectId or name, dir?, zip? |
design_preview |
Render a project's self-contained HTML to a full-page PNG for review — projectId or name, path?, dir?, width? |
design_get |
Read one file from a project — projectId, path |
design_status |
Report a project's chat/turn state — projectId |
design_edit |
Apply a direct file edit — projectId, path, edits |
design_delete |
Delete a project — projectId |
Setup
npm install # installs playwright-core (NO browser download — uses your Chrome)
node src/server.mjs login # opens Chrome once; log into claude.ai (session is then reused, invisibly)
Register as a local MCP (opencode example):
{ "mcp": { "claude-design": { "type": "local", "command": ["node", "/abs/path/claude.design-mcp/src/server.mjs"], "enabled": true } } }
CLI
node src/server.mjs login
node src/server.mjs list
node src/server.mjs create "minimal landing page for a coffee shop" coffee
node src/server.mjs iterate <projectId> "add a dark mode toggle to the header"
node src/server.mjs pull <projectId|name>
node src/server.mjs preview <projectId|name> [outDir] [width]
node src/server.mjs delete <projectId>
After the one-time login, list/create/iterate/pull run with no visible window
(off-screen Chrome) and reuse the persisted session.
Requirements
- Node.js 22+ (uses built-in
fetch/WebSocket;playwright-coreis the only npm dependency) - Google Chrome (the tools drive a dedicated Chrome profile)
- A claude.ai account with Design access (you log in once via
design_login)
Env
CLAUDE_DESIGN_PROFILE— dedicated Chrome profile dir (default~/.cache/claude-design-mcp/chrome-profile)CLAUDE_DESIGN_CHROME— path to Google Chrome (default: macOS Google Chrome)CLAUDE_DESIGN_CDP_PORT— remote-debugging port (default9377)CLAUDE_DESIGN_DIR— wheredesign_pull/design_previewwrite (default: the working folder)CLAUDE_DESIGN_HEADLESS— set1to drive headless Chrome instead of off-screenCLAUDE_DESIGN_TURN_TIMEOUT_MS— hard cap per generation turn (create ~420s, iterate ~300s defaults)CLAUDE_DESIGN_QUIET_MS— how long the turn network must stay silent before a generation is judged complete (default45000)
When is a generation "done"?
claude.ai/design drives generation as turns: your prompt streams in over a Chat RPC,
kept alive by RenewTurn keepalives (~every 10s) and ended by a ReleaseTurn. design_create /
design_iterate return once the files have settled AND the turn network has gone quiet for
CLAUDE_DESIGN_QUIET_MS — comfortably longer than the keepalive interval, so a generation is never
cut off mid-write (you always get a complete, coherent design, not a half-rendered one).
Note that claude.ai often runs an automatic refine pass that starts ~30s after the first design
settles, so the design keeps improving on the server after the tool has returned its first complete
version. To get the most-refined output, design_pull / design_preview always fetch the latest
state, or raise CLAUDE_DESIGN_QUIET_MS (e.g. 60000) to make create wait through later refine
passes (at the cost of a longer wait).
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.