Setup

Rams reviews the UI your coding agent writes. Connect it once, then ask your agent to review a component — you get issues back with severity, category, and exact file:line in about ten seconds.

Add the connector

Rams runs as a hosted MCP server at https://worker.rams.ai/mcp. You need a free Rams account first — sign up in under a minute, no payment, no CI to configure.

Claude (connector)

In Claude's settings, add a custom connector with the URL above. Claude opens a Rams page where you approve access to one workspace, then you're connected. No API key to paste.

Claude Code

  • claude mcp add --transport http rams https://worker.rams.ai/mcp
  • Then run /mcp, authenticate Rams, and approve access in the browser.

Cursor, VS Code, Codex, other agents

Any MCP-capable agent works. Add an HTTP MCP server pointing at https://worker.rams.ai/mcp and connect. Or paste this to your agent and let it set itself up:

Add the Rams MCP server to your MCP config — Streamable HTTP, url https://worker.rams.ai/mcp — and reconnect. It will open a browser to authorize; approve it. Then run a Rams quick_review on one UI file from this repo and show me what it found.

Use it

Ask your agent to review the files you just changed. Rams has three tools:

  • quick_review — a fast design check, about ten seconds, a fifth of a review credit. Run it constantly while you edit.
  • review_files — the deeper pass: a 0–100 score plus git-apply-able patches, for when you want something to keep.
  • verify_fixes — confirms your fixes landed. Free, never counts against your quota.

A good standing habit: tell your agent to run a quick_review after editing UI files, and a review_files before committing. Add that line to your project's CLAUDE.md or AGENTS.md and every session reviews before it ships.

What it costs

Every account includes reviews each month, free, shared across the connector and the GitHub App. Approving the connector lets it run reviews and spend that workspace's credits; it reads only the files you hand it and keeps none of your source. See Security and Privacy.

Questions? Email rams@rams.ai.