Connect Viberight
Viberight is an MCP server your agent uses to fetch rules, prompts and measurements.
Address: https://mcp.viberight.si/mcp
Have your key ready
You need a beta key. Store it as an environment variable, so it sits in no file you share. In your shell, for example in ~/.zshrc or ~/.bashrc:
macOS, Linux
export VIBERIGHT_KEY="YOUR_KEY"
On Windows in PowerShell:
PowerShell
[Environment]::SetEnvironmentVariable("VIBERIGHT_KEY", "YOUR_KEY", "User")
Connect
Claude Code
In the terminal, once for all projects:
Terminal
claude mcp add --transport http --scope user viberight https://mcp.viberight.si/mcp --header "Authorization: Bearer $VIBERIGHT_KEY"
Check that it is there:
Terminal
claude mcp list
Claude Desktop
Claude Desktop adds servers through a config file and needs the small bridge mcp-remote (Node 20 or newer) for servers with a key. Open the file: macOS ~/Library/Application Support/Claude/claude_desktop_config.json, Windows %APPDATA%\Claude\claude_desktop_config.json. Add the viberight block under mcpServers, put your key into AUTH_HEADER and restart Claude Desktop completely.
claude_desktop_config.json
{
"mcpServers": {
"viberight": {
"command": "npx",
"args": [
"-y",
"mcp-remote@0.15.2",
"https://mcp.viberight.si/mcp",
"--header",
"Authorization:${AUTH_HEADER}"
],
"env": {
"AUTH_HEADER": "Bearer YOUR_KEY"
}
}
}
}
Cursor
In ~/.cursor/mcp.json for all projects or .cursor/mcp.json in a project. Cursor reads the key from the environment variable:
mcp.json
{
"mcpServers": {
"viberight": {
"url": "https://mcp.viberight.si/mcp",
"headers": {
"Authorization": "Bearer ${env:VIBERIGHT_KEY}"
}
}
}
}
The first run
Tell your agent in plain words what to do. For example:
- “Audit https://my-app.com with Viberight, only the home page and the pricing page.”
- “Measure https://my-app.com with run_checks at 390 and 1440 pixels and show me the P1 findings.”
- “Fetch the polish prompt with these findings and fix security and broken things first.”
The six tools
- get_audit_prompt
- Fetches the audit prompt with the brief and the matching checklist.
- run_checks
- Measures a public address in a real browser, with axe for WCAG 2.2 AA and over HTTP.
- report_template
- Returns the findings format, so every finding is written the same way.
- get_polish_prompt
- Fetches the prompt for fixing, with the order and the full rules for your findings.
- list_rules
- Searches rules by area, severity, type or text.
- get_rule
- Shows one rule in full: rule, reason, check, severity.
Limits
- Per key 30 requests per minute, 500 requests and 20 measurements per day.
- Only public addresses over http and https on ports 80 and 443 are measured. Private networks, localhost and internal addresses are blocked, also after redirects.
- At most three widths and two languages per measurement. A run usually takes under two minutes.
- Remotely there are no checks on local paths and no signed-in pages. Your agent reads code and copy in the project itself.
- Screenshots and results do not stay on the server. We only log the host, the duration and the number of findings.
No key yet?
We hand out keys one by one during the beta.