Ask your analytics from Claude, Codex or any MCP client.

Connect your AI assistant to ClickClacks and ask in plain words. It reads your funnels, retention and traffic with nine read-only tools, and every answer says exactly what it was based on.

Your AI clientclickclacks

An illustration. The tools and arguments are real; the numbers are examples.

tools, every one of them read-only
9
for every client, over Streamable HTTP
1 URL
default window, or any start and end date
30 days
numbers guessed when data is unavailable
0

How it works

A key, a command, then a question.

There’s nothing to install on your side and no data to export. Your client talks to ClickClacks directly, with the key you give it.

  1. Create a key

    In Settings → Analytics API keys. It can only read, and it decides which project your client sees.

  2. Add the server

    One URL for every client. Settings → MCP has the command and a JSON config ready to copy.

  3. Ask in plain words

    Your assistant picks the tools and reads the numbers, and can show you the dates and filters behind every answer.

The tools

Nine tools. All of them read.

They cover the reports you already use in ClickClacks. Every tool takes an optional start and end date; leave them out and it reads the last 30 days.

Try asking

  • “Which funnel step loses the most people this week? Show the calculation.”
  • “Compare weekly retention for the last eight weeks. Stop if the response is degraded.”
  • “What are the top pages and referrers this month, and which friction findings overlap them?”
  • overviewOverview KPIs and the visitors seriessource, segment, property filters
  • event_breakdownsTop pages, referrers, countries and eventsdimension, plus report filters
  • list_eventsEvent rows, with exact-name search and pagingsearch, cursor, plus report filters
  • list_sourcesYour sources, their event counts and healthnone beyond dates
  • list_funnelsSaved funnels, their steps and summary metricsnone beyond dates
  • funnel_resultsOne saved funnel, with an optional property breakdownfunnelId, breakdown
  • retentionThe daily, weekly or monthly cohort gridinterval, plus report filters
  • list_segmentsSaved segments and how many people matchnone beyond dates
  • friction_findingsRage clicks, dead clicks and other findings seen in the windowstate: happening, stopped or muted

Report filters are a source, a saved segment or event-property filters, all resolved inside the key’s project. Full schemas

Privacy and safety

Read-only, scoped to one project, and honest about what it knows.

  • Read-only, by design

    All nine tools read. None of them creates, edits, deletes or sends anything, and the server tells your client so when it connects.

  • Scoped by the key

    The key decides which organization and project the client can read. Tool arguments can’t name another one: a request that tries is rejected.

  • Yours to switch off

    Keys carry a read-only analytics scope, are stored only as a hash and can be revoked in Settings at any time. Each key has its own rate limit.

  • Honest about gaps

    Every result says which dates, filters and project it covers. If a report can’t be read, it comes back empty and marked as degraded, never filled in.

every result carries

"provenance": {
  "window": {
    "start": "2026-08-25",
    "end": "2026-09-22"
  },
  "filters": { "segmentId": "seg_…" },
  "scope": {
    "organizationId": "org_…",
    "projectId": "prj_…"
  },
  "freshness": { "cache": "not-exposed" },
  "degraded": false,
  "reason": null
}

It comes back twice: as structured data for programs, and inside the text your assistant reads, so it can cite it.

The tools return what your ClickClacks reports already show, and nothing beyond them. That still means your AI client sees those results, so connect the clients you’d trust with a report, and give each one its own key.

Setup

Connected in about a minute.

Create a key, then paste one of these with cka_... replaced by it. Keep the key in your client’s secret settings, not in a prompt or a repository.

Server URL · Streamable HTTP, stateless, tools only

https://app.clickclacks.io/api/v1/mcp

Authorization · Bearer cka_live_…

Read the MCP guide

Claude Code

Run it in a terminal.

terminal
claude mcp add --transport http clickclacks https://app.clickclacks.io/api/v1/mcp --header "Authorization: Bearer cka_..."

Codex and other clients

For clients configured with JSON. Key names vary slightly between clients; the URL, transport and header don’t.

mcp config
{
  "mcpServers": {
    "clickclacks": {
      "type": "http",
      "url": "https://app.clickclacks.io/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer cka_..."
      }
    }
  }
}

FAQ

MCP, in detail.

  • Any MCP client that can reach a remote server over Streamable HTTP and send an Authorization header, such as Claude Code or Codex. Settings → MCP gives you the Claude Code command and a JSON config for the rest; key names vary a little between clients.

  • No. The server only offers nine read tools. There is no tool that creates, edits or deletes a funnel, a segment, a board or anything else.

  • The same reports you see in the app, for the one project the key belongs to: overview numbers, breakdowns, event rows, funnels, retention, segments, sources and friction findings. Whatever a tool returns goes to your AI client, so connect the ones you’d trust with a report.

  • Every tool takes an optional start and end date (YYYY-MM-DD), together or not at all. Without them it reads the last 30 days in UTC, including today.

  • The result is marked as degraded, with a reason, and the affected numbers come back empty. ClickClacks never hands your assistant placeholder or estimated figures dressed up as real ones.

  • Revoke its key in Settings → Analytics API keys. The next request from that client is refused. Give each client its own key, so you can switch one off without touching the others.

Start hearing your clicks.

One script tag. Your first report in minutes.

4.3 KB · pageviews, clicks and scroll depth, automatically