Lets AI agents see what's broken on a live website, from the outside: server errors, uptime, DNS, HTTPS, leaked files, broken pages, email setup, Cloudflare errors and recent changes, with evidence. Read-only.

Tools

list_sites: the websites in your Preventionlabs account, and whether each is verified.
run_report: a new full report on one verified site, as Markdown. Uses one of the site's daily runs.
latest_report: the most recent report, how long ago it ran, and runs left. Costs no run.

Only sites you've proven are yours can be checked, with the same daily limits as the dashboard.

Prompts

after-deploy: checks the live site after a deploy, fixes what broke, and runs the report again to prove it.
whats-wrong: finds whether a problem shows from the outside, and where the cause is.

In Claude Code: /mcp__preventionlabs__after-deploy.

How agents use it

The server tells every agent the loop: check, read the evidence, find the cause, fix it, then run the report again. Nothing counts as fixed until a new report shows it gone. Runs are limited (usually 2 per site per 24 hours), so agents spend them on a check and its re-check, and say so when a fix can't be proven yet. Past the daily runs, a report uses one of the account's report credits, if it has any.

Setup

Sign up at https://app.preventionlabs.com.au, add your site and prove it's yours.
In the dashboard, open Connections > AI agents and click Create API key.
Add the server to your agent.

Claude Code:

claude mcp add -s user preventionlabs -e PREVENTIONLABS_API_KEY=plk_your_key -- npx -y preventionlabs-mcp

Other MCP clients (JSON config):

{
  "mcpServers": {
    "preventionlabs": {
      "command": "npx",
      "args": ["-y", "preventionlabs-mcp"],
      "env": { "PREVENTIONLABS_API_KEY": "plk_your_key" }
    }
  }
}