MCP tools for coding agents

PageCanary's MCP server lets Claude Code, Cursor and other MCP clients check a deploy or a fix the way a visitor would.

Tool	What it does
`walk_flow`	The AI agent walks the site's main flow (signup, booking, checkout or contact form) in a real browser and returns the result, the steps and a recording. About a minute. Never sends, books or pays. Pass `plan` (steps in plain English, one per line) to walk your own steps instead.
`get_walk`	The result of a walk that was still running
`check_page`	One page checked for broken links and images, error messages, phone layout, the mobile menu, SSL and email setup, with how to fix each problem
`list_sites`	The sites your account monitors
`get_site_findings`	The problems PageCanary has confirmed on one of them, with evidence and fix steps
`add_site`	Start monitoring a site on your account: daily checks on desktop and phone and a weekly walk of its main flow. Counts toward your plan's sites (free 1, Starter 5, Pro unlimited).
`remove_site`	Stop monitoring a site on your account

Connect

Make an API key on API and integrations.

Claude Code:

claude mcp add --transport http pagecanary https://www.pagecanary.com/mcp --header "Authorization: Bearer pc_live_…"

Other clients that take a JSON config:

{
  "mcpServers": {
    "pagecanary": {
      "url": "https://www.pagecanary.com/mcp",
      "headers": { "Authorization": "Bearer pc_live_…" }
    }
  }
}

Try it

"Walk the signup on my preview deploy and tell me if it works."
"What has PageCanary found on example.com? Fix the top one, then check the page again."
"I just launched my-app.dev. Add it to PageCanary."
"On the preview, open pricing, pick Pro and check the checkout says $29 a month."

For a Vercel preview behind deployment protection, the agent can pass vercel_bypass_secret to walk_flow.