# Rams: design review for UI code, inside your agent

[Rams](https://www.rams.ai) reviews UI code the way a design-literate senior
engineer would: severity-rated findings with file and line, a 0–100 score,
and a concrete fix for each issue. It runs as a hosted
[MCP](https://modelcontextprotocol.io) server, so Claude Code, Cursor, Codex,
and any MCP-capable agent can call it while they work, before the code is
ever committed.

This repository is the connection guide. The engine is hosted. There is
nothing to install or run locally.

## Tools

| tool | what it does | cost |
|---|---|---|
| `quick_review` | Top issues on the files you just changed, ~10 seconds | 1/5 review credit (1/2 on the free plan) |
| `review_files` | Full scored review: 0–100, every issue with a fix, git-apply-able patches | 1 credit |
| `verify_fixes` | Confirms applied fixes actually landed | free |
| `review_url` | Reviews a live public page by its address, at desktop and phone width: top issues and a score, up to a minute, no patches | 1 credit |
| `usage` | Credits used and remaining | free |

The engine reviews React, Vue, Svelte, CSS, and SwiftUI against a rule set
spanning accessibility, hierarchy, color, type, spacing, motion, and
native-platform conventions. The live rule count is at
[rams.ai/rules](https://www.rams.ai/rules).

## Setup

Add the server and approve access in your browser. You sign in, or create a
free account (15 reviews to try it, no card), as part of that step. There is
no key to paste.

**Claude** · add Rams from the
[Claude connector directory](https://claude.ai/directory/rams), or add a
custom connector with `https://worker.rams.ai/mcp`.

**Claude Code**

```bash
claude mcp add --transport http rams https://worker.rams.ai/mcp
```

Then run `/mcp`, pick `rams`, and approve access in the browser.

**Cursor, VS Code, Codex, any MCP-capable agent** · add a Streamable HTTP
server at `https://worker.rams.ai/mcp` and connect. The agent opens a browser
to authorize. Step-by-step for each agent: [rams.ai/setup](https://www.rams.ai/setup).

### With an API key (CI, gateways, clients without OAuth)

Create a key at [rams.ai/app/settings](https://www.rams.ai/app/settings) and
send it as a Bearer header.

**Claude Code**

```bash
claude mcp add --transport http rams https://worker.rams.ai/mcp --header "Authorization: Bearer rams_YOUR_KEY"
```

**Cursor** · `.cursor/mcp.json`

```json
{
  "mcpServers": {
    "rams": {
      "url": "https://worker.rams.ai/mcp",
      "headers": { "Authorization": "Bearer rams_YOUR_KEY" }
    }
  }
}
```

**Codex** · `~/.codex/config.toml`

```toml
[mcp_servers.rams]
url = "https://worker.rams.ai/mcp"
http_headers = { "Authorization" = "Bearer rams_YOUR_KEY" }
```

Gateways that can't set headers can pass `?apiKey=rams_YOUR_KEY` instead.

Then ask your agent:

```
rams review my changes
```

## How it behaves

- Quick checks are built to run constantly while editing: after writing a
  component, after a refactor, before moving on.
- Full reviews are the commit-time artifact: the score you keep, the patches
  you apply.
- Criticals cap the score: one caps it at 59, two at 49, three or more
  at 39. A 90+ means something.
- Rams keeps the review (the score and each finding with its flagged lines),
  not your files. Details: [rams.ai/security](https://www.rams.ai/security).

## Elsewhere

- Web: [rams.ai](https://www.rams.ai) · Rules: [rams.ai/rules](https://www.rams.ai/rules) · Pricing: [rams.ai/pricing](https://www.rams.ai/pricing)
- Claude: [connector directory](https://claude.ai/directory/rams) · Official MCP registry: `ai.rams/rams` · [Glama](https://glama.ai/mcp/connectors/ai.rams.worker/rams) · [Smithery](https://smithery.ai/server/@eli-rousso/rams) · [mcp.so](https://mcp.so/servers/rams)
- GitHub App (reviews every PR automatically): [rams.ai](https://www.rams.ai)

## License

This connection guide is MIT licensed. The hosted engine is a commercial service by [HSLA0001 Inc.](https://www.rams.ai)