Cayu: the SEO tool I designed to work with verifiable data

What needed solving

Commercial tools give you a number and don't explain where it comes from. I needed to see the formula, cross two sources and know when a data point isn't reliable, because I was going to decide what to sell with those numbers.

The example that convinced me: «consultor seo» in Colombia. The commercial source said difficulty 0. The real top 10 had an average authority of 46, and they were consultants' personal sites, not empty pages.

keyword: "consultor seo" · ColombiaKD from the commercial source:  0Cayu Real-KD:                  42   (top 10 with average authority 46)→ The source underestimated: its link index for Colombia is thin.

What was done

Cayu is four MCP servers in Python (keywords, technical SEO, writing and design) that share a common library, and a React app to see them. Each server exposes its tools to an AI assistant, which is what runs them when I ask.

An instrument gets tested with two different inputs that must produce different outputs. When they produce the same thing, the instrument is lying. That's how I tested every tool before using it to decide anything, and that's how I found the two defects below.

How it turned out later

• 3 times

  commercial difficulty underestimated the real top 10

  Sep 14 and 17, 2026 · table below

• 2

  of its own defects found in use and fixed

  logged in DEFECTOS.md

• 0.18 USD

  to measure 31 keywords in two countries

  DataForSEO · Sep 17, 2026 · own ledger

• 0.25 USD

  to calculate Real-KD for 4 pillars in the US

  DataForSEO · Sep 17, 2026

• 5

  auditors flagged as suspect in the Sep 15 audit

  that's a result: the defects list is public

• A structured data auditor gave 100/100 to pages whose FAQs were hidden off-screen. It now detects off-screen text.

• The performance test reported a single PageSpeed run. It now runs several and reports the median and the range.

• «consultor seo» in Colombia and in Spain returned almost the same suggestion list, with Spanish cities in the Colombian one. Flagged as suspect and out of use until fixed.

What I learned, mistakes included

An instrument gets tested with two inputs that differ. When Colombia and Spain returned the same list, it wasn't data: it was a defect, and I had used it for two days without noticing.

The KD from commercial tools isn't used to decide. I confirmed it three times and I no longer argue about it: you measure the real top 10 or you don't measure.

And I learned to log defects in a visible file instead of fixing them quietly. An instrument without a list of known failures is an instrument I don't know what to distrust about.

If you want to check it yourself

Cayu isn't public yet. What you can check today, with any commercial tool: measure the difficulty of «consultor seo» in Colombia and then look at the authority of the first ten results. The gap between what the number says and what you see in the SERP is what Cayu calculates.

• The decisions that came out of this tool are in the keyword research case.

• The rendering and deployment auditors were used in the case about putting this site into production.

• What it audits in structured data is what I check in the AI SEO service (in Spanish).

What I can't claim

I don't claim Cayu is better than Semrush or Ahrefs: it does fewer things and it does them for me. I don't publish its code yet. And its difficulty is a formula of my own: it serves to compare keywords with each other, not to compare against another tool's KD.

It doesn't replace judgment either. When a source doesn't answer, it says so instead of inventing the number, and the decision is still mine.

Next measurement

The dates are written before the data exists. If I haven't updated this page by then, call me on it.