SEO Skill

SEO Skill

The SEO command for AI agents. A free Google Search Console MCP server and CLI: your agent reads your real Search Console, Google Analytics and Bing data, with Google update dates beside it, plus keyword research from your own DataForSEO or Ahrefs key.

Get started · Hosted MCP · Documentation · npm · Questions · Privacy · Security · License · Changelog

Ask your agent what is happening in search and get an answer built from your own Search Console, Bing Webmaster Tools, Google Analytics and Clicky data, with Google update dates laid over it. Add your own DataForSEO or Ahrefs key when you need keyword, SERP, competitor or backlink data.

The results are facts with their dates and limits. There is no advice baked in. Your agent reads the facts, shows you what matters and asks before it changes a page.

Install and run setup

You need Node 22 or newer.

npm i -g seo
seo start

seo start signs you in with Google. Then it offers to add an MCP server to the agents it finds (the free hosted server is the first choice, or you can run one locally) and to install the skill. That is all it asks. Nothing about your site is saved: each command names the site it works on.

List the properties your sign-in can read, then run the main report for one:

seo sites
seo report --site example.com

Connect the hosted MCP server

If you would rather not install anything, connect your agent to the free hosted server at https://seoskill.dev/mcp:

claude mcp add --transport http seoskill https://seoskill.dev/mcp
codex mcp add seoskill --url https://seoskill.dev/mcp

Sign in with Google when the client asks. That first Google account is your account, so there is no password or separate signup. The server keeps only your linked Google accounts and any API keys you save, encrypted. It keeps no report data, no Search Console or analytics rows and no request logs.

Your account page lets you add more Google accounts, save DataForSEO or Ahrefs keys, or delete everything. Each request uses the linked Google account that can read the site. The hosted MCP guide covers other clients, and the API keys guide shows how to add a key without pasting it into a chat.

A crawl fetches many pages, so seo crawl and seo internal-links run only on your machine, through the CLI or the local MCP server. Bing data also runs only there for now; when you ask the hosted server for Bing, it tells your agent which CLI command to run.

The eight commands

Command	MCP tool	What it returns
seo report	seo_report	The main report: an overview against the previous period, then every section's count and top rows
seo page --url <url>	seo_page	One URL: live page facts, Google index status and its queries
seo compare --date <date>	seo_compare	Equal periods before and after a date, with the Google updates in the window
seo index	seo_index	Pages that lost most of their clicks, checked live and with URL Inspection
seo research <action>	seo_research	Keyword metrics, ranked keywords, SERPs, competitors and backlinks from your key
seo data	seo_data	Raw Search Console rows, or Bing rows in the CLI and local server; the hosted tool also reads analytics landing pages
seo crawl <url>	seo_crawl	A site crawl with technical issues and the URLs they affect (local only)
seo internal-links --url <url>	seo_internal_links	Pages that could link to a target page (local only)

Each command takes its target (a site, URL or property) and an optional date window. --site takes a domain, a URL or a Search Console property, and seo sites lists the ones you can read. Commands that return rows also take --detail summary|rows|full and --offset. Brand terms, minimum impressions and the comparison period are set for you and shown in the output, so there is nothing else to tune.

Run seo help for the short list, or seo <command> --help for one command. The command docs explain each one.

Start with one report

Run the report without a section first:

seo report --site example.com

It shows site totals against the previous period, the pages and queries that moved, and a short count with top rows for each section. Then ask for the one section that matters:

seo report --site example.com --section quick-wins

The sections are overview, quick-wins, low-ctr, striking-distance, second-page, content-decay, cannibalization, traffic-drop and bing. Page sections check their top 20 URLs live and merge same-site redirects.

Each result ends with the command to run next. A traffic drop usually leads to seo compare with the date it started, then seo page for the URLs that moved:

seo compare --site example.com --date 2026-09-15
seo page --url https://example.com/pricing

Start with the summary. Ask for --detail rows --offset 20 for the next page of rows, or --detail full for every field, only when you need them.

Add keyword and competitor data

Research uses your own provider account, so each request costs your credits with that provider. Connect a key once:

seo providers dataforseo connect
seo providers ahrefs connect

On the hosted server, save the keys on your account page or send them as headers. Then ask for one action:

seo research keyword_metrics --keywords "invoice software,invoice template"
seo research serp --keywords "invoice software"
seo research ranked_keywords --target example.com
seo research competitors --target example.com
seo research backlinks --target example.com

Rows with a domain get a free Domain Rating column, credited "Domain Rating by Ahrefs". The hosted server adds it for everyone; the CLI adds it when you have saved an Ahrefs key. Provider numbers are estimates. Search Console stays the record of what your own site actually got.

Add Bing Webmaster Tools

Bing data is free. It works in the CLI and the local MCP server. Save a Bing Webmaster API key (Bing Webmaster Tools, then Settings, then API access):

seo providers bing connect

Agents and CI can set SEO_BING_API_KEY instead. seo providers bing status shows the connection and seo providers bing disconnect removes it. The hosted server has no Bing for now. When you ask it for Bing, it tells your agent to run the CLI instead, the same way it does for a crawl.

seo report then shows Google and Bing clicks and impressions for the same dates, and each engine's share of those clicks. The bing section puts Bing's top queries and pages beside Google's numbers for the same rows, and seo data --source bing returns the raw Bing rows. Bing and Google numbers sit side by side and are never added together. Bing lists only its top queries and pages by week, so a row missing from Bing's list is unknown, not zero.

Add your analytics

Google Analytics comes with your Google sign-in. List your properties, then pass the numeric property id to a Google Analytics command:

seo analytics google properties
seo ai-referrals --property 123456789

For Clicky, save your Clicky account login (every site) or one site's ID and sitekey on your account page. The hosted server and a CLI linked with seo login both read them:

seo data --source clicky --site example.com

A site saved on your machine with the Clicky provider package (seo analytics clicky connect --site-id 12345) works too, by its site ID. For any other analytics tool, build a provider package with the public provider SDK; the Clicky package is the example. Each source keeps its own metric names, so a Clicky visit is never relabelled as a Google Analytics user.

Use it with agents

The skill teaches your agent the eight commands, the one-report flow and the evidence rules. If you skipped it during seo start, add it now:

seo skill install

seo start can add the hosted or the local MCP server for you. To add the local MCP server later, to Codex, Claude Code, Claude Desktop or Cursor:

seo mcp install

Scripts can pick clients with --codex, --claude-code, --cursor or --claude-desktop. Other clients can start seo mcp serve as a local stdio server.

The local server and the hosted server serve the same tools: seo_sites, seo_report, seo_page, seo_compare, seo_index, seo_research and seo_data. The local server adds seo_crawl and seo_internal_links, plus seo_list_providers, seo_describe_provider and seo_run_provider for installed provider packages.

What the results can show

• Missing, partial, capped and filtered data is named. It never becomes a zero or an all-clear.
• Search Console, Bing, analytics, live page checks and provider estimates stay apart. Query rows do not add up to site totals.
• Flags such as likely automated queries and CTR gaps against the site's own curve are heuristics. They point at rows worth a look.
• A Google update in the same window is timing, not cause.
• An intentional noindex, canonical or redirect is an observation until you say it is wrong.

Nothing here promises clicks, rankings or indexing. Check a change by running the same command again over a later, complete window.

Upgrading from 0.3

Saved project profiles are gone. Every command takes --site (or --url), and --project or --client now stops with a note that says so. seo projects, seo client and the commands that saved history between runs (change-log, tests, content-groups, monitoring, schedule, index-watch, technical-watch, crawl-reports, crawl-diff and diagnose) were removed. Your agent keeps its own notes now. Each removed command names what to use instead. To stay on the old version, run npm i -g seo@0.2.41.

Upgrading from 0.2

Version 0.3.0 moved most reports into the eight commands. Old names such as seo quick-wins still run in 0.3 and print the new command to use. They stop working in 0.4.0. Some reports and commands were removed, including rank tracking, programmatic SEO, AI search readiness, llms.txt, Bing links, IndexNow, Semrush, local search, server logs and the monthly and narrative reports. Bing Webmaster search data came back in 0.3.0-beta.3. The old research reports became seo research. The changelog lists every old name and where it went.

Use it in CI and scripts

Add --json to any command. JSON mode never prompts. Pass --site every time.

seo report --site sc-domain:example.com --json
seo report --site sc-domain:example.com --section content-decay --json

A CI job can read Search Console with a Google service account. Give it the JSON key through your secret store and grant it access to the exact properties:

- name: Run the SEO report
  run: |
    npm i -g seo
    seo report --site sc-domain:example.com --json > seo-report.json
  env:
    SEO_GOOGLE_SERVICE_ACCOUNT_JSON: ${{ secrets.SEO_GOOGLE_SERVICE_ACCOUNT_JSON }}

Use it as a library

Install the same package in any Node 22 or newer project:

npm install seo

import { executeReport } from 'seo/mcp'
import { crawlSite } from 'seo'

const wins = await executeReport('quick-wins', {
  site: 'sc-domain:example.com',
})

const crawl = await crawlSite({
  url: 'https://example.com',
  maxPages: 100,
})

console.log(wins)
console.log(crawl.summary, crawl.issueGroups)

seo/provider-sdk is the public contract for provider packages. The Clicky provider is a working example.

Local data and Google access

Google tokens use your system keychain, with a private 0600 file fallback on headless machines. Settings, provider keys and a short cache of Google responses stay in your local config directory. Reports and crawls are not saved: each run returns its results and your agent decides what to keep. A command still reaches the network when it needs Google, your site or a provider you connected.

seo privacy
seo doctor
seo auth logout --all
seo reset

The hosted server keeps only your linked sign-ins and saved keys, encrypted, and no report data or logs. Read the privacy policy for every network boundary and the telemetry page for the anonymous usage counts and how to turn them off.

Documentation

Questions, bug reports and feature requests go through GitHub Issues. Report suspected vulnerabilities privately through the process in SECURITY.md.

• Getting started
• Hosted MCP
• API keys
• Your hosted account
• CLI
• Commands
• What is stored
• Privacy policy
• Public usage stats
• Terms of use
• Contributing
• Trademark and brand policy

Develop locally

Most users do not need the source checkout. If you want to contribute:

git clone https://github.com/iannuttall/seo.git
cd seo
pnpm install
pnpm build
node dist/cli.js start --dry-run

Run the full quality gate before opening a pull request:

pnpm build
pnpm typecheck
pnpm test
pnpm lint
pnpm pack --dry-run

Contributor architecture and report rules live in AGENTS.md.

License

The code is available under Apache-2.0. Project names and artwork are covered separately by the trademark and brand policy.