SuperAPI AI Directory

webflow-insights

Data & analytics

View repository

Updated 10-18 01:51

About this skill

Webflow Insights

Webflow's MCP server reads three of its products: Analyze, Optimize and AEO Analytics. Not every site has all three. Many have Analyze only, some have none, and AEO Analytics is mostly found on enterprise plans.

This skill checks which products a site has, runs the right reads, interprets the results honestly and answers in plain language. A product the site doesn't have is treated as a normal state, never as an error.

Hard rules

  1. Read-only. Use only data_sites_tool (list_sites, get_site), the read actions of data_pages_tool and data_cms_tool, data_analyze_tool, data_optimize_tool and data_aeo_tool. Never publish, edit, create or delete anything.
  2. A missing product is not an error. Never show a status code, an error name or raw JSON, and never describe a product as blocked, forbidden, failed or denied. Use the wording in references/availability.md.
  3. Never invent a number. Report only fields that came back. Label any figure you calculated, such as a conversion rate or a percentage change, and say what it's calculated from.
  4. A crawler visit is not a citation. Webflow says so too. Citations and mentions come only from the prompt-tracking reports.
  5. Don't call a test early. No winner below 0.95 significance on an A/B test, and none at all for personalisation.
  6. Always state the date window you used, in the site's local dates.

Before you start

  1. Find the site. data_sites_tool > list_sites. If there are several live sites and the person hasn't said which, list them by name and domain and ask once. Then call get_site for the site's timeZone.
  2. Check what's available. Run the light probes in references/availability.md for the products the question needs. A check-up probes all three. Record each as available, not on the plan, not set up, or no access, and reuse the result for the rest of the conversation.
  3. Resolve the date window in the site's time zone. See references/analyze.md > Date windows.

Four ways in

Request What to do
No question, or an open one ("how are we doing?") Deliver a snapshot of the last 28 days straight away, then suggest three or four questions to ask next. Don't open with questions; the only one allowed first is which site. Layout in references/reports.md
A question ("which pages get the most traffic?") Run the fewest reads that answer it, using the reference for that product
A monthly check-up Last complete month against the one before, across every available product. Layout in references/reports.md
Diagnosing a change ("why did traffic drop?") Confirm the change, find where it happened, check what changed on the site, then conclude with a confidence level. Steps in references/reports.md

Answer format

  1. The answer, in one or two sentences, first.
  2. The numbers behind it, with the date window.
  3. What to do next, up to three points, each tied to a number.
  4. What the data can't tell you, when it matters to the question.
  5. One availability line, only when a missing product is relevant.
  6. Two or three questions to ask next, built from what the data showed and using the site's real page, goal and test names. Most people won't know what to ask, so this is how they find the next useful question. If they reply "the first one", run it.

Use plain words ("visits", not "sessions"), give counts alongside percentages, and leave out ids, field names and raw JSON.

Reference Documentation

File Read when
references/availability.md Every run, before the first real query. Detecting each product and the exact wording for anything missing
references/analyze.md Traffic, pages, sources, goals, engagement, scroll depth or page flow
references/optimize.md Tests, experiments, variations or personalisation
references/aeo.md AI crawlers, AI search, tracked prompts, visibility or citations
references/reports.md A snapshot, a check-up, a diagnosis, or a question that spans products. Also holds the answer layouts and the bank of suggested questions

Requirements

  • The Webflow MCP server connected to your AI agent, authorised for the site's workspace.
  • Access in Webflow to whichever of Analyze, Optimize and AEO Analytics the site has. The agent can only see what the person using it can see.

License

MIT