Connect
Connect Claude Code to runcurve MCP.
Add the remote MCP server with one command, approve OAuth, then ask Claude Code to build a small web app and return its live URL. The studio is there when you want to inspect or edit the files yourself.
Tested path: Claude Code
Run this in a terminal with Claude Code installed. The command adds runcurve as a user-scoped remote HTTP server, so it is available in your Claude Code projects.
claude mcp add --transport http --scope user runcurve https://mcp.runcurve.com/mcp- 1. Add runcurve. Run the command above.
- 2. Authorize it. Run this command, or open /mcp inside Claude Code and choose runcurve's authentication option.
claude mcp login runcurve - 3. Complete the browser handoff. Claude Code opens your browser to sign in to runcurve and approve the connection. After approval, it returns to Claude Code automatically. On a headless machine, use the command below, open the printed URL on your local browser, then paste the full callback URL back into the terminal when prompted.
claude mcp login --no-browser runcurve
Make one safe first riff
In a Claude Code chat, paste this prompt. It creates only a public greeting page; no secrets, outside network access, or visitor data are involved.
Use runcurve to create a public riff named "Hello runcurve". Make its home page say "Hello from my first runcurve riff." Return the working public URL.Claude Code should return a unique public *.run.runcurve.com URL. Open it and you should see “Hello from my first runcurve riff.” Every later save is live at the same URL.
What this supports
This is the published setup path for Claude Code's remote HTTP MCP connection with OAuth. A raw server URL is not plug-and-play in every assistant. Use a client that explicitly supports remote Streamable HTTP MCP and OAuth with dynamic client registration; other clients are added here only after a full connection test.
After connection, the build-a-riff Agent Skill is the runtime reference for your assistant. This page is the first-run guide.