Microsoft Clarity MCP Server

Nhận phân tích hành vi cho các thuộc tính đã cấu hình của bạn, bao gồm lưu lượng truy cập, mức độ tương tác, độ sâu cuộn và tín hiệu khó chịu

GitHubDùng thử MCP nàyĐược tài trợ

Tài liệu

@urdigital/mcp-server-clarity

An MCP (Model Context Protocol) server exposing Microsoft Clarity's behavioral analytics to Claude, Claude Code, and any other MCP-compatible client.

Install

No install needed — run directly with npx:

   npx -y @urdigital/mcp-server-clarity

Configure

Get a token from your Clarity project → Settings → Data Export → Generate new API token. Tokens are scoped to a single project — you need a separate token per project if you're tracking multiple sites.

Add to your MCP client config (e.g. Claude Desktop's claude_desktop_config.json):

   {
     "mcpServers": {
       "clarity": {
         "command": "npx",
         "args": ["-y", "@urdigital/mcp-server-clarity"],
         "env": { "CLARITY_API_TOKEN": "your-project-token" }
       }
     }
   }

Tools

ToolDescription
clarity_get_insightsBehavioral analytics for the configured project: traffic, engagement, scroll depth, and frustration signals (rage clicks, dead clicks, quickback clicks, script errors)

Important behavior, confirmed by testing against a real project

Clarity's Data Export API is a single, narrow endpoint with real limitations worth knowing before you build on it:

  • Only the last 1-3 days of data are retrievable at all. There is no way to query older data through this API.
  • No dimensions vs. dimensions is a trade-off, not additive. Calling with no dimension1/2/3 returns a broad 16-category snapshot (traffic, engagement, frustration metrics, plus Browser/Device/OS/Country/PageTitle/ ReferrerUrl/PopularPages breakdowns). Specifying any dimension swaps that out entirely for a narrower metric set (frustration + engagement + traffic only), cross-tabulated by whichever dimension(s) you gave. You get depth on one axis or breadth across many — not both in the same call.
  • numOfDays is validated; dimension names are not. An out-of-range numOfDays (anything but 1, 2, or 3) returns a hard HTTP 400 with an empty body — no error message. An invalid dimension name, by contrast, returns HTTP 200 and silently falls back to the default no-dimension snapshot instead of erroring. This tool restricts dimension inputs to a strict enum of the valid values for exactly this reason — it's the only real safeguard against a silently wrong result, since Clarity itself won't tell you.
  • Small daily request quota per project (Clarity doesn't expose your exact remaining count anywhere in the API response) — choose numOfDays and dimensions deliberately rather than probing repeatedly.

License

MIT