From a sentence to a diagram.
Arketo connects to the assistant you already build with over MCP: the open standard that lets an AI assistant use tools outside itself (what MCP is, in plain words). You describe the system, it draws and audits it. Here is the whole loop.
Five steps, start to report
From an empty canvas to a finished report, this is the whole loop the assistant runs with you.
1 · Create your free account
Free scans to start, no card. You get an isolated workspace and a private URL for your assistant, shared with no other account, and a canvas that persists between sessions. Subscribe later, once it has earned it.
2 · Connect your assistant
Arketo is a standard MCP server, so it drops into any MCP-capable client: Claude, Cursor, Windsurf, VS Code, Codex, and more. In Claude Code it's one command.
claude mcp add --transport http arketo "<your-mcp-url>" --header "Authorization: Bearer <your-key>"3 · Describe your system
Tell it in plain language: a React frontend, an API gateway, a Go backend, Postgres, a Redis cache. Components appear as you talk and the connections route themselves.
4 · Ask it to analyze
Say "find the risks." It walks the graph and marks bottlenecks, security gaps, and scaling limits by severity, right on the components they affect.
5 · Download the report
Ask for a write-up and it produces a full Markdown report with a phased scaling plan. Drop it into your repo, paste it into a ticket, or hand it to a teammate.
Wire up the tool you already use
The same URL in every case: the private one your dashboard hands you. Only where you paste it changes.
I build with Claude Code
One command in your project terminal. Claude Code registers the server and finds the tools on your next message.
claude mcp add --transport http arketo "<your-mcp-url>" --header "Authorization: Bearer <your-key>"I build with Cursor
Drop it into .cursor/mcp.json in your project, or ~/.cursor/mcp.json to have it in every project, then reload.
{
"mcpServers": {
"arketo": { "url": "<your-mcp-url>" }
}
}Another MCP client
Windsurf, VS Code, Codex, Claude Desktop and the rest all take a URL. Paste yours wherever that client keeps its MCP servers.
<your-mcp-url><your-mcp-url> is a placeholder. Your dashboard prints each of these with your own URL and key already filled in, so setup is a copy and a paste.
It draws while you talk.
This is a faithful miniature of the real canvas: components laid into client, service and data tiers, edges routed and labelled between them, the request path marked as it carries the load, and flagged components carrying their severity badges. On your own canvas it builds like this while you describe it, a piece at a time.
You talk to it the way you already do.
No diagram syntax, no schema to learn. These are the kinds of sentences that drive the canvas.
We have a Next.js app talking to a Node API, Postgres, and S3 for uploads. Draw it.
Add a Redis cache in front of the database and a worker queue for the email jobs.
Now find anything that will break at 50x the traffic and write it up.
Wire it up in one command.
Get your canvas and MCP URL, paste it into your assistant, and start describing. The first diagram takes minutes.