v1.0.0 to v1.1.1

+ version: "1.1.1"

+ # Optional, when Cogny MCP is connected (all read-only)

+ - mcp__cogny__refresh_mcp_discovery

+ - mcp__cogny__list_integrations

+ - mcp__cogny__mcp__Microsoft_Clarity__get_live_insights

+ - mcp__cogny__mcp__Microsoft_Clarity__get_insights_history

+ - mcp__cogny__mcp__Microsoft_Clarity__get_api_quota

+ - mcp__cogny__mcp__Google_Analytics_4__tool_list_account_summaries

+ - mcp__cogny__mcp__Google_Analytics_4__tool_run_report

+ - mcp__cogny__mcp__Google_Analytics_4__tool_list_conversion_events

+ - mcp__cogny__mcp__Plausible__breakdown

+ - mcp__cogny__mcp__Plausible__query

+ - mcp__cogny__mcp__GrowthBook__list_experiments

+ - mcp__cogny__mcp__GrowthBook__get_experiment_results

+ ## Works with

+

+ The review runs on the public page with zero accounts. Connected analytics turn opinions into measured problems.

+

+ | Job | Primary | Also answers it | No account |

+ |-----|---------|-----------------|------------|

+ | Conversion rate of this page, by device and source | GA4 (`tool_run_report`) | Plausible (`breakdown`, `query`) | — (score from heuristics) |

+ | Where visitors get stuck (scroll, rage/dead clicks, quickbacks) | Microsoft Clarity (`get_live_insights`) | — | — |

+ | What was already A/B tested here | GrowthBook (`list_experiments`, `get_experiment_results`) | — | Ask the user |

+

+ If GA4 and Plausible are both connected, show both rates side by side; they count sessions and visitors differently.

+

+ From the fetched HTML you can check:

+ - Mobile viewport meta tag (`curl -sL <url> | grep -i 'name="viewport"'`; WebFetch can't see `<head>`)

+ - Content readability on small screens (font sizes and line lengths in the CSS, if inline)

+ Touch target sizes and whether the CTA sits above the fold need a rendered page. WebFetch and curl can't measure them. Run the URL through PageSpeed Insights (https://pagespeed.web.dev, Lighthouse flags "Touch targets" / "tap targets"), or open it in a browser's device mode at 375px wide. If neither is possible, mark both "not measured (needs a browser)" instead of scoring them.

+

+ ### 5a. Measure load time

+ Step 7 scores load speed, so measure it rather than guess:

+

+ ```bash

+ curl -sL -o /dev/null -w "ttfb=%{time_starttransfer}s total=%{time_total}s html=%{size_download}B redirects=%{num_redirects}\n" "<url>"

+ # Rough page weight: HTML + the scripts, styles and images it references

+ curl -sL "<url>" | grep -oE '(src|href)="[^"]+\.(js|css|png|jpe?g|webp|avif|svg)[^"]*"' | head -40

+ ```

+

+ Fetch the largest few assets with the same `-w "%{size_download}"` call and sum them. Report "TTFB X s, HTML Y KB, ~Z KB total across N requests (curl, no JS execution)". Over 1 s TTFB or ~3 MB total weight costs load-speed points. For field data (real users), use `/cwv-audit`.

+

+ ### 5b. Measured behavior (enhanced when connected)

+ If Cogny isn't connected, skip this and note what connecting adds: real conversion rates by device and source (GA4/Plausible), where visitors get stuck (Clarity), and what was already tested (GrowthBook).

+

+ Call `mcp__cogny__list_integrations` once and skip whatever isn't connected.

+

+ - **GA4:** `mcp__cogny__mcp__Google_Analytics_4__tool_run_report` with `property_id`, `date_ranges` (last 28 days), `dimensions: ["landingPagePlusQueryString", "deviceCategory", "sessionSourceMedium"]`, `metrics: ["sessions", "engagementRate", "keyEvents", "sessionKeyEventRate"]`, filtered to the page path. Check `tool_list_conversion_events` so you know which key events count.

+ - **Plausible:** `mcp__cogny__mcp__Plausible__breakdown` (`site_id`, `dimension: "visit:entry_page"`, `metrics: ["visitors", "bounce_rate", "conversion_rate"]`, `date_range: "30d"`). Conversion rate needs a goal filter, e.g. `filters: [["is", "event:goal", ["Signup"]]]`.

+ - **Clarity:** call `mcp__cogny__mcp__Microsoft_Clarity__get_api_quota` first (10 requests/day), then `get_live_insights` with `num_of_days: 3`, `dimensions: ["URL", "Device"]`, `metrics: ["Scroll Depth", "Rage Click Count", "Dead Click Count", "Quickback Click"]`. Dead clicks on a non-link element or a scroll depth that ends above the CTA are direct evidence for a top-5 change. Heatmaps and recordings aren't in the API; tell the user where to look in Clarity.

+ - **GrowthBook:** `mcp__cogny__mcp__GrowthBook__list_experiments`, then `get_experiment_results` for any experiment on this page. Don't recommend a variant that already lost.

+

+ Mobile vs desktop gaps from this data take priority over heuristic scores in step 7.

+

+ - Load speed: [score] — TTFB [X]s, ~[Y] KB, [N] requests (curl)

+

+ Measured (if connected; name the source on every number):

+ [If nothing is connected: "Measured: not connected. Connecting GA4/Plausible, Clarity or GrowthBook adds real conversion rates, friction data and past test results."]

+ - Conversion rate: [X]% desktop / [Y]% mobile (GA4 or Plausible)

+ - Clarity: [N] rage clicks, [N] dead clicks on [element], avg scroll depth [X]%

+ - Already tested: [experiment, result] (GrowthBook)