Skip to content

The methodology

The MCP serves the AI SEO method as well as your data: how an agent looks up a process, fetches it, and scores a passage before you publish it.

Updated October 20, 2026

An agent that can read a dashboard still needs to know what to do about what it reads. The connection serves the method as well as the numbers: processes, methodologies and skills, fetched on demand.

Look up, then fetch

Two tools, used as a pair.

Look up takes a question in natural language, something like "content triage" or "internal linking strategy", and returns a ranked list of matches: an id, a type, a name, a description and a score. No file content, so it is cheap and the agent can scan the options.

Fetch takes the ids it chose and returns the bodies in one bundle, optionally pulling in what those documents depend on. One round trip for a whole workflow rather than one call per file.

The intended pattern is one look up and one fetch, even when a job needs several documents.

What is in there

Three types, and a search can be narrowed to any of them.

  • Methodologies. How something works and why, the reasoning a decision rests on.
  • Processes. The steps for a job, in order, with what to do at each.
  • Skills. Narrower techniques a process calls on.

The practical effect is that "audit this site's AI visibility" becomes a defined sequence the agent follows rather than whatever it improvises, and two different sessions produce comparable work.

Scoring a passage before you publish

chunkscore takes a query and a set of passages and scores them. It is the same scorer behind retrieval readiness, so an agent can check a draft against the query it is meant to serve while the draft is still a draft.

That closes the loop that otherwise takes a week: write, publish, crawl, score, discover the passage does not answer the query, rewrite. Scoring the draft moves the whole cycle into one session.

Using it well

In practice that is one prompt:

We are invisible on comparison questions. Find the right process for that
and plan the work against our actual data.

Two instructions worth giving your agent once:

  • Look it up before answering. For anything about diagnosis, content strategy, technical fixes or link building, the method exists and is better than a general heuristic.
  • Fetch the process, then follow it. Reading a process and then doing something else is a familiar failure, and asking for the step it is on keeps it honest.

The methodology is shared across every member and is kept current centrally, so a change to the method reaches your agent without you updating anything. Your project's own rules are a separate thing and belong in project files.

Questions about the methodology

Do I have to install anything?
No. The methodology is served over the same connection as your data and is always current, because it is fetched rather than bundled.
Is it available on a read-only connection?
Yes. Looking up and fetching the method are reads, so every plan has it.
Can I add our own processes?
Put them in project files as a rules registry. The served methodology is the shared method; your project's specifics belong with your project.

Related

Was this page helpful?

Not tracking yet?

Seven days free, on your own model key. Point it at one brand and read what comes back.

Start your free trial