Free and open source
Ask Claude Code, Cursor, or Codex for three directions. They land on a canvas side by side, you pick one, and the agent builds it in your codebase. All on the plan you already pay for.
No credits, no card, no server key.
Works withClaude Code
CursorCodex
opencodeand any MCP client.
One prompt, three answers
Your agent writes each screen as a page and puts it on the canvas while you watch.
How it works
Tell your agent what the screen is for and ask for three layouts. It writes each one as a self-contained HTML page and calls Wirely to put it on your canvas.
Frames fill in as each page lands. Pan, zoom, open the compare view, and look at the three next to each other instead of scrolling a chat for screenshots.
Pick one element on the canvas and copy its link. The agent edits that element and leaves the rest of the page alone. Reviewers can pin comments the same way.
The agent reads the page it wrote, so the build matches the design instead of guessing from a screenshot.
>
The tool names are the ones the MCP server exposes.
Pricing
No tiers, no credits, no cap on agent calls. Your agent runs on the subscription you already have, and pages render in a sandbox.
No agent? Bring a key:Google AI StudioOpenRouterZ.aiUnsplash
$0forever
Connect your agentSetup is one line in your agent. The docs cover Claude Code, Cursor, Codex, and opencode.
FAQ
Something missing? The docs go deeper, and GitHub issues reach the maintainer.
No. There are no plans, credits, or limits on agent calls. Your agent runs on the subscription you already pay for, and Wirely only stores the pages it writes.
Claude Code, Cursor, Codex, and opencode have setup steps in the docs. Any other client that supports remote MCP servers over streamable HTTP works too.
Not if you use an agent. A key is only needed to generate pages inside Wirely without an agent. You save it in settings, and Wirely encrypts it before storing it.
No. Every page renders in a sandboxed iframe without access to the Wirely origin, under a strict Content Security Policy that only allows a short list of script and image hosts.
You tell your agent which page you picked. It reads the HTML it wrote and builds the screen in your project. It can then capture your running app and get back what differs from the design.
Only you, unless you send a review link. People you send it to sign in, see the pages, and pin comments on them.
Yes. The source is MIT licensed. You need Bun, a Postgres database, and a Clerk app. The self-hosting guide in the docs lists every step.
Add one line to your agent and ask it for three directions.