Traditional analytics track clicks and pageviews. They have no concept of an AI agent invoking your WebMCP tools, completing a multi-step workflow, and driving a transaction — all without a browser session.
WebMCP Analytics is designed to close that gap: tool-level telemetry, agent traffic attribution, and schema diagnostics — built on the protocol itself.
One script tag · Full visibility in 5 min · No credit card
Illustrative example — how agent analytics differs from traditional tracking
example-travel.com · agent session #a8f2csearch-flights{from:"SFO", to:"JFK"}234msselect-flight{flightId:"UA-2847"}89mscheckoutrequestUserInteraction → approved1.2sconfirmtransaction complete156msSubmitEvent.agentInvoked + toolactivatedWhy existing analytics fail
Chrome 146 shipped WebMCP. AI agents now interact through structured tool invocations — not page loads, not clicks, not scrolls. Every analytics platform you use was built for humans.
An AI agent that books a $450 flight generates zero page views, zero scroll events, zero session duration in Google Analytics. But it just made you $450.
Without protocol-native analytics, that revenue is invisible — attributed to nothing, optimized by no one.
The SDK
The WebMCP Analytics SDK (< 5KB) auto-hooks into navigator.modelContext to capture every tool registration, invocation, result, and error.
quick-start.html1<"text-cyan-300/70">class="text-blue-400/80">span "text-cyan-300/70">class="text-white/20 italic">"text-cyan-300/70">class="text-blue-400/80">span>2<"text-cyan-300/70">class="text-blue-400/80">script "text-cyan-300/70">src="https://cdn.web-mcp.net/analytics.js"3 "text-cyan-300/70">data-site-id="ar_site_abc123">"text-cyan-300/70">class="text-blue-400/80">script>registerTool()Tool registrations and unregistrationsexecute()Invocations with parameters and timingreturnSuccess/error results and response typesagentInvokedAgent detection on form submissionstoolactivatedTool lifecycle DOM eventsprovideContext()Context state changesrequestUserInteraction()User approval flows and outcomesProtocol-native dashboard
PreviewEvery metric maps to a real WebMCP protocol event — tool invocations, success rates, schema mismatches, and agent identification via SubmitEvent.agentInvoked.
example-travel.comSAMPLE DATATool Calls
24h window
Agent Share
Success Rate
Investigate -0.4pp
search-flightsselect-flightcheckoutget-seat-mapcancel-bookingsearch-flights94.2%select-flight97.1%checkout88.7%get-seat-map99.4%cancel-booking96.8%search-flights✓params: {from:"SFO", to:"NRT"}234mscheckout!requestUserInteraction → approved1.2sselect-flight✓params: {flightId:"JL-0001"}89mssearch-flights✗schema mismatch: 'date' expected ISO—checkout✗toolcancel — user denied—Illustrative data · Protocol-native telemetry via registerTool() instrumentation and SubmitEvent.agentInvoked detection
Traffic Intelligence
“How much of our business is coming through AI agents?” Without WebMCP, you cannot answer this. With it, you can answer it in real time.
Agent share grew from 3.2% → 22.4% in 12 months. At this rate, agents will represent 40%+ of interactions by Q4.
Agent traffic behaves differently from human traffic. Agents invoke tools in rapid succession, complete transactions in seconds, and leave no scroll or click data. Understanding these patterns requires protocol-level telemetry.
Comparison metrics are illustrative of expected behavioral differences
Revenue Attribution
PlannedWhen an AI agent completes a multi-step tool chain that ends in a transaction, you need to know. Revenue attribution connects tool invocations to business outcomes so you can answer the question: how much revenue did agents generate?
search-flightsinvokedselect-flightinvokedcheckoutuser approvedconfirmcompletedsearch → select → checkout → confirmComplete purchase journeysearch → select (abandoned)Partial journeyWhy this matters: AI agents interact with your site through tool invocations, not pageviews. Traditional analytics attribute a transaction to the last click — but when an agent invokes checkout after a multi-step tool chain, there is no click to track. Revenue attribution closes that gap.
Revenue attribution is a planned feature. Join the waitlist for early access.
GA4 tracks pageviews. WebMCP Analytics tracks which tools agents invoke, which parameters they send, and which schema mismatches cause failures. One script tag. Data flowing in 5 minutes.