Canaryusers

What canaryusers MCP server does

The canaryusers MCP server gives an AI coding assistant access to CanaryUsers’ hosted UX testing service. It accepts a live or preview URL, has AI personas inspect the deployed experience, and returns findings intended to guide implementation work. The service focuses on problems such as confusing interfaces, broken mobile layouts, dead-end conversion paths, and visual glitches.

Results are more actionable than a single quality score: scans return findings with explanations and concrete fixes. This makes the server relevant after deploying a UI change, before launching a product, or when investigating friction in signup, checkout, onboarding, pricing, or landing-page experiences.

How it works

The server is remote rather than a package that you clone and run locally. MCP clients send requests to https://www.canaryusers.ai/api/mcp and authenticate with a CanaryUsers bearer token.

canary_scan accepts a deployed URL and a scan depth. Quick is the default and performs a fast static check without consuming a credit. Deep renders the page, examines desktop and mobile presentation, and follows important flows such as signup, checkout, or onboarding. Deep scans take approximately 60–90 seconds and use credits.

The canaryusers MCP server is designed to inspect the deployed result, not application source code. After a scan, an assistant can use the returned findings directly or retrieve the complete report in Markdown for review or further processing.

Setup and configuration

Create or obtain a token from the CanaryUsers dashboard by opening CI & API, enabling MCP, and copying the token. Configure the remote MCP server in a client that supports HTTP MCP connections:

{
  "mcpServers": {
    "canaryusers": {
      "url": "https://www.canaryusers.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_CANARYUSERS_TOKEN"
      }
    }
  }
}

The same endpoint can be added through Claude Code with its HTTP transport and an authorization header. The repository documents configuration examples for Cursor, Windsurf, VS Code, and Claude Desktop as well. No local service process is described as necessary.

Tools and capabilities

• canary_scan: Starts a UX scan for a deployed URL. Its depth option supports quick and deep; quick is the default and free, while deep includes visual and flow testing and uses a credit.
• list_recent_scans: Lists recent scans with the scan ID, URL, score, grade, status, and date.
• get_report_markdown: Retrieves the full Markdown report for a scan ID.

Together, these tools support a workflow of scanning a deployment, checking its status or history, and obtaining a report that can be read by an AI assistant or incorporated into developer documentation.

Limitations and notes

The target must be a deployed web app URL; the material does not describe source-code analysis or a local-only scan mode. Quick scans are free, but deep scans consume credits. The repository does not specify fixed plan limits because pricing and monthly allotments can change, so current details should be checked on the CanaryUsers pricing page.

The MIT license applies to the metadata and documentation repository. The hosted CanaryUsers product is separate and closed source. The canaryusers MCP server therefore cannot be run by cloning this repository alone; access depends on the hosted endpoint and a valid token.