Claude Code for SEO: Live Keyword, SERP and Backlink Data With treg

By Ammar Ahamed, Founder, Cocoon · 3 November 2026 · Facts checked against treg.to, its docs and its GitHub repository on 3 November 2026

What Claude Code can do for SEO, and what it can't see

Point Claude Code at a website repo and it will happily do the work most SEO audits end with: shorten titles that truncate, add missing meta descriptions, write BlogPosting and FAQPage structured data, fix broken internal links, generate a sitemap and compress images. Open-source skills make this even more thorough. When we searched Google for "claude code seo", the top two organic results were the same free audit plugin, claude-seo, and its site.

But an audit skill gives your agent a method, not data. Ask it "how many people search for this?", "who ranks on page one right now?" or "which sites link to our competitor and not to us?" and it has two choices: say it doesn't know, or guess. A guessed search volume reads exactly like a real one, which is why it's dangerous. The fix is a data layer the agent can call. That is the gap treg fills: the skill brings the method, treg brings the numbers.

What people search for: "Claude Code SEO" and its neighbours

To show the workflow rather than describe it, we did this post's keyword research with Claude Code and treg. One call to DataForSEO's Google Ads volume endpoint priced 70 keywords at once. A few of the results:

Our data: Google Ads monthly averages, all locations, English, via DataForSEO through treg, 3 November 2026. One call, 70 keywords, $0.09.

Two things stand out. People search for the plumbing far more than the use case: "claude code mcp" gets about seventeen times the searches of "claude code seo". And the vendor-specific servers ("semrush mcp", "ahrefs mcp") already draw nearly 2,000 searches a month each, with high ad prices, which tells you SEO teams are actively shopping for a way to get their data into an agent.

The SEO data treg gives Claude Code

Here is the SEO part of the catalog as listed on 3 November 2026. Where several providers do the same job, treg shows them side by side with price, measured success rate and speed, and a routed tool can pick for you.

Note the shape of the pricing. Serpstat bills per keyword it returns (and bills a minimum even when it returns nothing), SE Ranking bills per keyword too, and DataForSEO bills per call regardless of how many keywords are in it. For a long list of niche, long-tail terms, one DataForSEO call is usually the safer buy. For a handful of head terms, Serpstat is cheaper. Your agent can read this off catalog get and choose.

Set up Claude Code for SEO with treg

In Claude Code, paste:

set up treg - https://treg.to/llms.txt

or run the installer yourself:

curl -fsSL https://treg.to/install.sh | sh
treg login
treg catalog search "keyword search volume"

Then connect Google Search Console and GA4 through OAuth from your treg dashboard, so your agent can read your own clicks, impressions and positions. Those calls run on your account and aren't metered. New verified accounts start with $1.00 of credit, which at the prices above is a lot of keyword research.

What we ran, and what it cost

Three calls, made through treg's MCP tools from Claude Code on 3 November 2026:

The results page was the most useful call for writing. The AI Overview for "claude code seo" described three things: audit plugins installed with /plugin or npx skills add, Claude Code fixing meta tags, sitemaps and structured data in the codebase, and scripts that pull Search Console and GA4 data into files the agent can query. The related searches ("Claude Code SEO agent", "Claude Code SEO skill", "Claude Code SEO plugin") told us which sub-topics this post should cover. That's a content brief for less than a fiftieth of a cent.

8 Claude Code SEO workflows, with prompts you can copy

Each prompt asks the agent to show the cost first. Keep that habit: it turns a surprise bill into a two-second check.

1. Price a keyword list before you write anything

Why it works: one bulk call is cheaper than many small ones, and the "flag under 50" line stops you writing for terms nobody searches.

2. Build a brief from the live results page

3. Check whether AI answers cite you

Answer engines are now a second front page. A weekly snapshot of who gets cited is the simplest way to track it.

4. Find backlink gaps against a competitor

5. Work your "striking distance" queries from Search Console

This is where Claude Code is at its best: the data comes from treg, and the edits land in your codebase in the same session.

6. Track rankings weekly without a rank-tracker seat

7. Price programmatic pages before you build them

Pricing 120 combinations costs cents. Building 120 pages nobody searches for costs weeks, and risks Google treating them as scaled, low-value content.

8. Fix the technical basics in the repo (no data needed)

Not every SEO job needs paid data. Run this one first; it's free and often the biggest win.

Semrush MCP, Ahrefs MCP, DataForSEO MCP or treg?

If you searched for "semrush mcp" or "ahrefs mcp", you've probably noticed both vendors now ship official MCP servers. They're good, and they suit a different situation from treg:

Our view: if your team opens Semrush or Ahrefs every day, keep it and use the vendor's MCP server. If you're a smaller team, an agency running occasional audits, or you need SEO data alongside social and enrichment data, treg is the lighter choice. And they combine: connect your Semrush key to treg and your agent reaches Semrush through the same token, unmetered.

Guardrails for an SEO agent

• Cap the spend. Give the agent its own treg token with a spend cap (treg org agent-new --cap), and ask for the expected total before any loop. A routed call already stops at $1 by default.
• Check big decisions twice. Keyword tools disagree. Before you commit a quarter's content plan to a term, check its volume with a second provider.
• Keep a person on "publish". Let the agent research, brief and draft. A person reads before anything goes live. Google's spam policies target pages produced at scale mainly to rank, however they were made.
• Save raw answers. Ask the agent to write each response to a file. You can re-read it later without paying twice, and you'll catch parsing mistakes.
• Mind the terms. Data from a provider comes with that provider's terms. Don't republish raw third-party datasets as your own.

The on-page side still matters as much as the data. If you want a ranked list of the AI SEO tools worth trying beyond Claude Code, see our best AI SEO tools guide. If outbound is your next job after organic, the same setup powers the workflow in GTM Engineering With Claude Code.

Frequently asked questions

Sources

treg

• treg.to homepage and catalog
• The treg skill: operating guide

Other MCP servers

• Understanding Semrush MCP, Semrush Knowledge Base
• Getting started with Ahrefs MCP, Ahrefs Help Center
• DataForSEO MCP server, DataForSEO

SEO

• claude-seo, an open-source SEO plugin for Claude Code
• Spam policies for Google web search, Google Search Central