Open source SEO audit: 398 checks, a CLI, an app and a skill

I'm the founder and CEO of SEOmator, where we're building an AI SEO agent for B2B SaaS. SEOmator's SEO audit is open source. On 19 November 2026 we released version 8 under the MIT licence: 398 checks in 20 categories that run on your own machine, from a command line, a desktop app or a Claude Code skill.

Try it in two commands: npm install -g @seomator/seo-audit, then seomator audit https://yoursite.com. You don't need an account or an API key.

The code has been public in the SEOmator SEO Audit repository since January 2026, and OpenAlternative lists it as an open-source alternative to Semrush, Ahrefs, Sitebulb and Lumar. I started running SEOmator in October, and version 8 is the release where every screen opens the same way: the score first, then one list of fixes ranked by how many points each one wins back.

Is there an open-source SEO audit tool?

Yes, and there's more than one open-source SEO audit tool on GitHub. When I searched "seo audit open source" on 20 November 2026, four of Google's US top 10 were open-source SEO audit tools under the MIT licence: OpenSEO, SEOnaut, Claude SEO and ours.

None of their pages puts them side by side, so I did. SEOmator is one of the products I run, so you know which row is mine. Every cell comes from the tool's own page or repository, read on 20 November 2026.

ToolLicenceHow you run itWhat the report gives you
SEOmator SEO AuditMITA Node.js command-line tool, a desktop app, a local dashboard, or a Claude Code skill and MCP server398 checks in 20 categories; fixes ranked by the score points each returns
OpenSEOMITA hosted app, or self-hosted33 issue types: 4 critical, 20 warnings, 9 info
SEOnautMITA self-hosted Go app with a MySQL database, in DockerIssues in three levels: critical, high and low
Claude SEOMITInside Claude Code, with Python 3.11 to 3.14Up to 17 specialists per audit, scored across 7 weighted areas

The MIT licence lets you use, copy, change and ship the code "free of charge" and "without restriction", in the words of the Open Source Initiative's MIT text.

They're good at different things. OpenSEO calls itself "the open source alternative to Ahrefs and Semrush" and bundles keyword research, backlinks and rank tracking with its audit, at up to 50 pages per audit on the free plan and up to 10,000 on a paid one. Claude SEO runs as agents inside Claude Code and weights content quality (23%) and technical SEO (22%) highest; it had 31,004 GitHub stars on 20 November. SEOnaut is a self-hosted crawler with 913 commits behind it. Ours is built to run unattended: a plain command-line tool with CI exit codes, and an engine an AI agent can call.

What does the SEO audit check?

398 checks in 20 categories. Like most open-source audits, it covers titles, meta descriptions, headings, broken links, redirects and canonical tags. The rarer checks are the ones that need a real browser.

With --cwv, the audit renders each page in your own Chrome, Chromium or Edge and measures Core Web Vitals against Google's Web Vitals guide: LCP within 2.5 seconds, CLS of 0.1 or less. INP needs someone to interact with the page, so it's measured only with --simulate-interaction, and reported as synthetic and unscored. The same render runs the axe-core accessibility engine, which says it finds "on average 57% of WCAG issues automatically". For AI search readiness, it checks that citation crawlers can reach the page and that its structured data matches what the page shows.

Here are the 20 categories, with each one's rule count and weight in the score (README at version 8.0.1):

CategoryRulesWeight
Core2711%
Performance3010%
Links278%
Images158%
Security268%
Technical SEO197%
Accessibility467%
Crawlability395%
Structured data195%
JavaScript rendering175%
Content275%
Social133%
E-E-A-T163%
URL structure143%
Redirects113%
Mobile132%
Internationalization132%
HTML validation122%
AI/GEO readiness132%
Legal compliance11%

Accessibility has the most rules (46), but Core weighs the most (11%): titles, meta descriptions, canonical tags, robots directives and H1s.

The CLI's category breakdown for books.toscrape.com, a public demo site. Recorded for the 8.0 release on 19 November 2026.

How are the fixes ranked?

By the score points each fix returns, so a warning on every page can come before a failure on one.

Most of the tools I read rank by severity, with labels like critical, high and low. Points add up to exactly what your score is missing, so you can see what the first five fixes are worth before you start. On the books.toscrape.com demo, the desktop app's fix queue opens with "The first five return 3.8 of the 14.8 points this site is missing".

The rule I care about more than any single check is what happens when a check can't take a reading. The audit says "not measured" and gives the reason, such as "needs a browser render". It doesn't quietly count it as a pass.

How do you run it: CLI, desktop app or Claude Code skill?

All three run on your own machine with Node.js and share one engine and one local history. Six of the eight results I read for this search say you host an open-source audit yourself, and three of them name Docker. This one needs neither Docker nor an account.

From the command line

The @seomator/seo-audit package installs the SEO audit CLI as a seomator command:

npm install -g @seomator/seo-audit
seomator audit https://example.com
seomator audit https://example.com --crawl --max-pages 50 --cwv
seomator audit https://example.com --format html -o report.html

Reports come out as console text, JSON, HTML, Markdown or LLM-ready XML (--format llm), which you can pipe straight into a model.

For CI, the exit code tells you what happened: 0 means the audit passed, 1 means a gate failed, and 2 means the run couldn't happen. That way a CI job can tell a bad page from a broken pipeline. --fail-under sets the score that fails the build, 70 by default, and a rule set to fail_ci in seomator.toml fails it on its own.

In the desktop app

The desktop app opens on the score, then the fix queue, with a strip called "Where the points went" that shows what each category costs you. Every audit you run stays in a local history, per site.

The desktop app on books.toscrape.com: the fix queue, then "Where the points went", 14.8 points missing across 17 categories.

There's a catch: no installer yet. You run the desktop app from source (clone the repository, then npm install, npm run rebuild:electron and npm run electron:dev). If that's more than you want, seomator serve gives you the same screens in your browser at 127.0.0.1:7360, with no build step. It binds to your own machine only and makes no outbound requests apart from the audits you start.

Inside Claude Code

A Claude Code skill is a SKILL.md file of instructions; in Anthropic's words, "Skills extend what Claude can do". Add the SEO audit skill to Claude Code with npx skills add seo-skills/seo-audit-skill, then ask in plain words: "Audit https://mysite.com and tell me what to fix first."

It also runs as an MCP server, so an agent can start audits and read the findings itself. MCP is, as its documentation puts it, "an open-source standard for connecting AI applications to external systems".

claude mcp add seomator-audit -- npx -y @seomator/seo-audit mcp

How did the launch go on X and LinkedIn?

Most of it happened on X. I posted at 08:11 UTC on 19 November, about 18 minutes after 8.0.0 reached npm. By 08:57 UTC the next morning the post had 42.7K views, 21 replies, 63 reposts, 670 likes and 1K bookmarks. More bookmarks than likes, which I read as people saving it to try later.

GitHub moved with it. GitHub's starring API returns a timestamp for every star, so we counted them by day:

Day (UTC)New stars
13 Nov 20261
14 Nov 20260
15 Nov 20260
16 Nov 20263
17 Nov 20260
18 Nov 20262
19 Nov 2026 (launch)119
20 Nov 2026, to 08:5633

That's 152 new stars between the X post and 08:56 UTC the next day, against 48 in the 31 days before launch, about 1.5 a day. It took the repository to 824. It wasn't the first wave, though: 157 stars came in on 14 to 16 October.

152 new GitHub stars between the X post and the next morning, against 48 in the 31 days before it.

It's still small next to my markdown.new launch, whose X post passed 960K views in February. My guess at why: markdown.new worked the second you pasted a URL, as MyPreferredSources.com did the second you typed a domain. An audit asks you to install something first.

LinkedIn was quieter, with 18 reactions and 3 comments:

What doesn't the audit do yet?

The first question under the LinkedIn post was a fair one: "does it handle big sites with 10k+ pages okay?" Not really. A crawl is sized for an audit, 10 pages by default and up to 1,000 per run on the command line, so a 10,000-page site needs an SEO crawler built for that size. Another commenter had good advice here: start with a few page templates, because a fix to a template "often fixes hundreds of pages at once".

Up to 1,000 pages per run on the command line. Audit your templates first, not every URL.

Does it read Search Console data?

No. "Search Console data is not part of the audit yet," as the README says. Five of the eight results I read bring Search Console in, and one comment on a Reddit thread about an open-source sitemap audit names the gap: "the open source tools on pypi cover basics but miss the search console cross referencing which is where the real value is". The data is already there through Google's Search Console API, and it's the gap I'd close next.

Does it track rankings, keywords or backlinks?

No. Six of the eight results bundle rank tracking, keyword and backlink data with the audit, usually bought from a data provider. This audit checks only what it can fetch and render, so it needs no paid data and no key. Tracking belongs to the SEOmator platform at seomator.com, whose keyword and AI-prompt tracking has a waitlist today. I'd rather keep the audit narrow. Popupsmart V2 cost us 1.5 years of adding features before I learned that lesson.

FAQ

Can ChatGPT do an SEO audit?

It can explain and rank findings well once it has measurements. Getting those means rendering pages in a browser, timing them and reading every tag across a crawl, which a chat model needs a tool for. Pipe seomator audit https://yoursite.com --format llm into the model, or let Claude Code call the audit through the skill or the MCP server, and it works from real numbers instead of guesses.

Is there a free online SEO audit tool available?

Yes. SEOmator's free SEO audit tool runs in your browser with no signup. It renders one page and runs 172 on-page checks across 16 weighted categories in about 30 seconds. Use it for a quick look at one page, and the CLI when you want to audit a whole section of a site or check it on every deploy.

How do I do an SEO audit?

Run a tool against your site, fix what it finds in order of impact, then run it again. With this one: npm install -g @seomator/seo-audit, then seomator audit https://yoursite.com --crawl --max-pages 50, adding --cwv for Core Web Vitals. Work down the fix queue from the top and re-run to confirm each fix landed. My Popupsmart SEO case study covers the rest of SEO, from site structure to keyword research and content.

Is an open-source SEO audit really free?

An open-source SEO audit is free in the licence sense: MIT lets you use, change and redistribute the code at no charge. What can cost money is what a tool runs on, like a server if you self-host or the keyword and backlink data some tools buy. This one needs no paid data, account or API key, so the only cost is the time a crawl takes.

Does my site's data leave my machine?

Not unless you send it somewhere. History stays in a local SQLite database at ~/.seomator/audits.db, and seomator serve binds to your own machine only. A public site you audit can't point the audit at your private network, with one exception for now: in a --render browser render, the browser loads the page's own resources without that check. Pipe a report into a model, though, and the findings go to that model's provider.

How the numbers were counted

The X and LinkedIn counts are what each public post showed at 08:57 UTC on 20 November 2026, not the platforms' own analytics. GitHub stars are counted by the date each was given, from GitHub's API at 08:56 UTC the same day; a star is someone bookmarking the repository, not a user of the tool. Rule counts and weights come from the README at version 8.0.1.

The quickest test is your own site: npm install -g @seomator/seo-audit, then seomator audit https://yoursite.com.

All writing
  1. Emre Elbeyoglu speaking at Fuckup Nights Ankara in front of a slide of the WIRED article Robert F. Kennedy Jr.'s Microsoft-Powered Chatbot Is Back Online

    · 4 min read

    LiveChatAI and the Kennedy Campaign: The OpenAI Rule We Didn't Know About

  2. Mert Özeren, Oğuzhan Ulukaya and Emre Elbeyoglu in front of the neon ISTANBUL Slush'D sign at Istanbul Ignited by Slush'D 2026

    · 2 min read

    Istanbul Slush'D 2026: Introducing SEOmator and Notes on Agentic Commerce

  3. Emre Elbeyoglu and Nisa Selin Kosova at the microphones for Startup Talks #1 by Istanbul Ignited by Slush'D, under the title Pahalı Bir Ders

    · 3 min read

    Popupsmart V2: More Features Cost Us 1.5 Years and $1M