Claude SEO skill review: verdict, safety and alternatives
Yes, Claude has SEO skills, and the most-starred is AgriciDaniel's Claude SEO plugin for Claude Code. It bundles 26 sub-skills and 19 agents that run scripted site audits. We rate v2.4.1 Recommended with caveats, 70/100, with safety grade C. The maintainer fixes reported bugs fast, but audit findings need checking before you act on them.
- Skill
- Claude SEO (AgriciDaniel/claude-seo)
- Verdict
- Recommended with caveats
- Score
- 70/100
- Confidence
- Moderate
- Safety grade
- C
- Best for
- Claude Code users who want scripted multi-agent site audits and will verify Critical findings
- Poor fit
- Teams forwarding audit numbers to clients unchecked; Windows users wanting zero-friction setup
- Reviewed version
- v2.4.1 · commit ff87fce
- Review date
- 2026-09-30
- Works in (verified)
- Claude Code
- Works in (claimed)
- claude.ai hosted plugins; Codex (unmerged community PR)
Get the skill: claude-seo repository on GitHub · SKILL.md at the reviewed commit · skills.sh listing
Is there an SEO skill for Claude?
Yes, there are several SEO skills for Claude. The most-starred is AgriciDaniel's Claude SEO plugin, which we rate 70/100. Other options include a different "Claude SEO Skill" by Alex Labat, plus InfraNodus and SE Ranking skills.
An Agent Skill is a folder whose SKILL.md file tells an AI agent how to do one job. Our guide to what counts as an Agent Skill covers the format in full. "Claude SEO skill" names a category, not one product.
| Option | Maker | What it is | What it needs | Our verdict |
|---|---|---|---|---|
| Claude SEO | AgriciDaniel (Daniel Agrici) | Plugin with 26 sub-skills | Python, Chromium, optional API keys | Recommended with caveats, 70/100 |
| Claude SEO Skill | Alex Labat (claudeseoskill.com) | Single skill with DOCX and PDF reports | ZIP upload to Claude | Not evaluated |
| InfraNodus SEO skill | InfraNodus | Search intent and content-gap graphs | InfraNodus MCP server and a free account | Not evaluated |
| Hawk Academy SEO skills | StudioHawk | 56 downloadable skills | None stated | Not evaluated |
| SE Ranking SEO skills | SE Ranking | 7 skills | SE Ranking MCP server and a paid plan | Not evaluated |
Alex Labat's Claude SEO Skill is a separate product with its own audit format. This page reviews AgriciDaniel's plugin, the one at github.com/AgriciDaniel/claude-seo.
What is Claude SEO by AgriciDaniel?
Claude SEO by AgriciDaniel is an open-source Claude Code plugin with 26 SEO sub-skills and 19 specialist agents. It runs site audits, schema, GEO, local and Google API checks through bundled Python scripts.
Claude Code is Anthropic's terminal coding agent, and it loads skills and plugins from disk. Our list of Claude Code skills by verdict ranks the others we reviewed. Claude SEO's job is to audit a site's technical, on-page, schema, content and AI-search readiness. Its output is a list of prioritised fixes. GEO means generative engine optimisation, the work of getting cited in AI answers.
An orchestrator skill routes each request to a sub-skill. Most checks call bundled Python scripts. Its users are SEO consultants, agency staff and developers on Claude Code.
| Attribute | Value |
|---|---|
| Publisher | Daniel Agrici (AgriciDaniel), not Anthropic |
| Type | Claude Code plugin bundling skills, agents and a hook |
| Reviewed version | v2.4.1, commit ff87fce |
| Current release | v2.4.2, released 2026-10-04 |
| Licence | MIT |
| Files at review | 446: 35 SKILL.md, 19 agents, 202 references, 169 scripts |
| Official site | claude-seo.md (publisher-owned) |
| Install line | /plugin marketplace add AgriciDaniel/claude-seo |
The counts differ by source. The README says 26 sub-skills, 19 specialist agents and 34 commands. Parts of claude-seo.md say 25, 18 and 30. The agenticskills.io directory lists 33 sub-skills. This page uses the README counts.
Is the Claude SEO skill worth installing?
Yes, the Claude SEO skill is worth installing when you verify its findings. It scores 70/100, Recommended with caveats, at Moderate confidence. Most reported bugs get fixed within days, but its scripts and APIs keep breaking in new places.
Our page on how verdict scores work explains each measure.
| Measure | Score | Why |
|---|---|---|
| Effectiveness | 70 | Scripted checks with real audits (#177, #143); held down by open #322 and #330 |
| Reliability | 60 | Lost findings, false GSC totals, render and encoding bugs; most fixed, three open |
| Safety | 72 | SSRF #110 and key leak #122 fixed; credentials needed and the hook runs globally |
| Efficiency | 60 | Heavy instructions and multi-agent fan-out; single commands load sub-skills on demand |
| Compatibility | 62 | Built for Claude Code; hosted upload failed until v2.2.6; Windows bugs recur |
| Maintainability | 88 | 28 releases in 12 months and 28 contributors; outside PRs usually re-implemented |
| Setup friction | 55 | Plugin install, Python venv, Chromium, Node and per-service API keys (#137) |
| Audit finding accuracy | 62 | False zero clicks (#130) and empty pages scoring 90 (#210) fixed; #330 open |
| Large-site audit completion | 70 | #177 wrote zero findings on 1,755 URLs; v2.3.0 fix not independently re-tested |
| Third-party API currency | 55 | Bing, Moz and Keywords Everywhere fixed; #322 open; #319 flags DataForSEO v3 drift |
| Install and runtime provisioning | 62 | Early plugin installs were non-functional (#137, #139); v2.2.4 added setup and doctor |
| Instruction weight vs value | 60 | About 314 KB of SKILL.md text plus 19 agents; hub routing limits idle cost |
| Side effects outside SEO work | 55 | validate-schema.py runs on every Edit/Write with no path filter (#321) |
Three questions have no published answer yet. No one has checked a live audit of a known site against ground truth. The token cost of a full audit is unmeasured, and so is the v2.3.0 fix past 1,000 URLs. Compare Claude SEO with other skills ranked by verdict before you choose.
The best fit is a developer or SEO practitioner on Claude Code who checks Critical findings before acting. Two groups fit poorly. One is teams that need audit numbers they can forward to clients unchecked. The other is Windows users wanting a zero-friction install, or anyone unwilling to run Python and Playwright.
What changed since our review?
Since our review, Claude SEO v2.4.2 shipped on 4 October 2026. It changed the schema hook's error output and added an optional seo-cockpit plugin. Bugs #322, #321 and #330 remain open.
The verdict above describes v2.4.1, commit ff87fce. These changes reached GitHub after our 30 September review:
- v2.4.2 (2026-10-04) adds the optional seo-cockpit 0.3.3 companion plugin. It needs Claude Code 2.1.287 or later.
- seo-cockpit brings a spend guard on paid API calls, plus /seo-cockpit, /seo-spend and /seo-doctor commands.
- Google APIs can fall back to your own gcloud sign-in, set with CLAUDE_SEO_GOOGLE_AUTH=adc.
- The schema hook now sends critical errors to stderr with exit code 2, so Claude sees them. Warnings arrive as additional context.
- The DataForSEO cost table was corrected, to $0.103.
- Issue #321 stays open on 2026-10-21, because the hook still fires on every edit.
- Issues #322, #330 and #319 stay open on 2026-10-21.
The release notes call the hook change a fix that is part of #321. The issue itself remains open. The score stays at 70 for v2.4.1.
How many tokens does a Claude SEO audit use?
A full Claude SEO audit of a 50-page site uses an estimated 400,000 input and 40,000 output tokens. Parallel subagents drive that cost. The one SKILL.md file we scanned measures about 1,104 tokens per load.
The token figures below each measure something different.
| Figure | Value | Label |
|---|---|---|
| Full /seo audit of a 50-page site | 400,000 input, 40,000 output tokens | Our estimate for a typical task |
| Typical call with the skill file loaded | 402,104 input, 40,000 output tokens | Estimate plus the file load |
| Scanned SKILL.md (4,415 bytes) | 1,104 tokens | Measured file size per load |
| All 35 SKILL.md files in the bundle | About 314 KB | Measured; loaded on demand |
| Audits per week | 2 | Our estimate for a typical task |
The 1,104-token figure covers one file, skills/seo-plan/SKILL.md. The hub loads sub-skills only when a command needs them, so idle cost stays low. A full audit is a different case. It fans out to many parallel subagents, and each one reads its own instructions and pages.
Some costs sit outside the token count. DataForSEO, Ahrefs, Firecrawl and Google Cloud NLP calls bill separately on your own accounts. Chromium and a Python venv add disk use and setup time. No one has measured a full audit's real token cost, so the audit figures stay estimates.
Is the Claude SEO skill safe to install?
Yes, with care: the Claude SEO skill is safe to install at safety grade C. It ships about 60 Python scripts, a hook that runs on every edit, and installers. Many features need Google, Moz, Bing or DataForSEO credentials.
Grade C means the skill needs credentials or broad permissions. Read how we grade skill safety before you hand API keys to any SEO plugin. Our file review found these 10 facts:
- Endpoints: scripts call Google PageSpeed, CrUX, GSC, GA4, NLP and Indexing, plus Moz and Bing Webmaster.
- Data sources: scripts reach Common Crawl, IndexNow and DataForSEO endpoints too.
- Credentials: scripts read a Google API key or OAuth, AHREFS_TOKEN, FIRECRAWL_API_KEY, MATOMO_TOKEN and INDEXNOW_KEY.
- Runtime: Python 3 runs in a managed venv, with Playwright Chromium for rendering and Node for the hook.
- Hook: a PostToolUse hook runs validate-schema.py on every Edit or Write, with no path filter.
- Network guards: fetchers go through url_safety.py, with SSRF and DNS-rebinding guards added after report #110.
- Untrusted content: content-ingesting agents carry an untrusted-content rule since v2.3.0 (#291).
- Models: agent files pin their models, 14 to Sonnet and 5 to Opus.
- Extensions: optional installers fetch MCP servers from npm and GitHub.
- Disclosure: the maintainer enabled private vulnerability reporting after #289.
The safety measure scores 72/100. Two serious reports were fixed in tagged releases. One was an SSRF bypass in URL validation (#110). The other was pagespeed_check.py printing the Google API key in error output (#122). The hook is the widest exposure. It runs in every project where the plugin is enabled, whether or not you are doing SEO work.
Keep API keys in a scoped config, and disable the plugin in projects that never need it. Grant paid-service keys only for the extensions you plan to use.
What does the community report about the Claude SEO skill?
The community reports real audits with the Claude SEO skill: 15 evidence items, 9 strong. A 1,755-URL audit wrote zero findings (#177), and GSC totals showed false zero clicks (#130). Both were fixed.
Our evidence comes from GitHub issues and pull requests dated 2026-05 to 2026-09.
| Issue | What was reported | Date | Strength |
|---|---|---|---|
| #177 | Full audit on a 1,755-URL site wrote zero findings files; fixed in v2.3.0 | 2026-07 | Strong |
| #130 | GSC totals summed query rows, producing false zero-click reports; fixed in v2.2.4 | 2026-06 | Strong |
| #137 | Marketplace install did not provision Python deps or Chromium; setup command added | 2026-06 | Strong |
| #122 | pagespeed_check.py printed the Google API key in error output; fixed | 2026-06 | Strong |
| #110 | SSRF bypass in URL validation reported and fixed with pinned safe fetchers | 2026-05 | Strong |
| #291 | Only 1 of 14 content-ingesting agents treated pages as untrusted; fixed in v2.3.0 | 2026-09 | Strong |
| #312 | Keywords Everywhere client hit a removed endpoint (404); rewritten in v2.4.0 | 2026-09 | Strong |
| #322 | NLP entity extraction always fails with HTTP 400; open at the pinned commit | 2026-09 | Strong |
| #321 | Schema hook fires on every edit; its stdout errors never reach the model | 2026-09 | Strong |
| #143 | Reported schema false negative not reproduced on v2.2.0 in three runs | 2026-06 | Weak |
| #330 | parse_html drops spaces around inline tags in headings and link text | 2026-09 | Weak |
| #252 | Windows local audit left orphaned find processes at full CPU; partly fixed | 2026-08 | Weak |
| #199 | claude.ai hosted plugin upload rejected the bin/ directory; fixed in v2.2.6 | 2026-07 | Weak |
| #289 | No private security channel existed; maintainer enabled private reporting | 2026-09 | Weak |
| PR #198 | Outside fix PR closed unmerged; maintainer re-implemented it in v2.2.6 | 2026-07 | Weak |
Users run the plugin on real sites and file detailed bugs. The maintainer answers most reports with a named fix version. Outside pull requests are usually closed and re-implemented rather than merged.
What are the Claude SEO skill's known failures?
The Claude SEO skill has three open known failures. Google NLP entity extraction always fails with HTTP 400 (#322). The schema hook fires on every edit (#321), and headings lose spaces (#330).
The worst failure for this job is an audit that reports confident but false findings. Wrong zero-click totals or missing schema lead a site owner to make the wrong changes. Two such cases, #130 and #210, were fixed before our pinned commit. These are the failures still tracked, with their states on 2026-10-21:
- NLP entities (#322): open. Google Cloud NLP entity extraction always returns HTTP 400, confirmed in nlp_analyze.py.
- Schema hook (#321): open. v2.4.2 changed only where the hook's errors go; it still fires on every edit.
- Heading text (#330): open. parse_html drops spaces around inline tags in headings and link text.
- DataForSEO (#319): open. The bundled client predates the DataForSEO v3 migration.
- Windows processes (#252): partly fixed. Local audits left orphaned find processes at full CPU.
Windows users meet a wider set of path, cp1252 encoding and process bugs (#186, #114, #290). Each integration is one more place for an outside API to change under the plugin. That is why third-party API currency scores 55.
Does the Claude SEO skill work in Codex and other agents?
No, not in Codex today. The Claude SEO skill is verified only in Claude Code, and claude.ai hosted plugin upload works since v2.2.6. Codex support exists only as an unmerged community pull request.
| Agent | Status | Source |
|---|---|---|
| Claude Code | Verified | Our file review at ff87fce |
| claude.ai hosted plugins | Claimed; upload fixed in v2.2.6 | #199, #298 |
| Codex | Claimed; community PR unmerged | Repository pull requests |
| Codex SEO port | Separate project named in the README | Not evaluated |
The plugin relies on Claude Code features: slash commands, subagents and a PostToolUse hook. Hosted plugins in the Claude app rejected the bin/ directory until v2.2.6. See our list of skills in the Claude app for options built for claude.ai. The README names a separate Codex SEO port, which we have not evaluated.
How do you install the Claude SEO skill?
Install the Claude SEO skill in Claude Code with /plugin marketplace add AgriciDaniel/claude-seo, then /plugin install claude-seo@agricidaniel-claude-seo. Run /seo setup next, which builds an isolated Python runtime and Chromium for the audit scripts.
The README (checked 2026-10-21) gives two install routes. The plugin route is the default. The script route clones the repository and runs an installer.
How do you install it as a Claude Code plugin?
Install Claude SEO as a Claude Code plugin with two slash commands, then run setup. The doctor command confirms that the runtime works.
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo
/seo setup
/seo doctor
- Add the marketplace with the first command.
- Install the plugin with the second command.
- Run /seo setup to build the Python venv and download Playwright Chromium.
- Run /seo doctor to confirm the scripts can run.
Setup exists because early marketplace installs did not provision Python deps or Chromium (#137). Our guide to installing skills in Claude Code covers plugin folders and scopes. The README mentions an AI Marketing Hub Pro variant; we have not evaluated it.
How do you install it with the install script?
Install Claude SEO with the install script by cloning the repository, then running install.sh on macOS or Linux. Windows users run install.ps1 after the clone.
git clone --depth 1 https://github.com/AgriciDaniel/claude-seo.git
bash claude-seo/install.sh
The README warns against piping a remote script into PowerShell with irm | iex. Clone first, read the script, then run install.ps1 from the local copy.
A third route exists. The skills.sh listing offers npx skills add agricidaniel/claude-seo. That listing does not mention the Python runtime the scripts need. The README route provisions it, so we treat the npx route as incomplete.
How do you run a Claude SEO audit?
Run a Claude SEO audit with /seo audit followed by a URL. It sends parallel agents and writes a scored report. Start with /seo page on one URL you know, and check each finding against GSC.
The README lists 34 commands. These are the ones a first audit uses:
| Command | What it does |
|---|---|
| /seo setup | Builds the Python venv and Chromium |
| /seo doctor | Checks that the runtime works |
| /seo audit |
Runs a full multi-agent site audit with a scored report |
| /seo page |
Audits one page |
| /seo technical | Runs technical checks |
| /seo content | Reviews content quality |
| /seo schema | Detects, validates and generates JSON-LD |
| /seo geo | Checks AI-search readiness (claimed-only) |
| /seo local | Runs local SEO and Maps analysis (claimed-only) |
| /seo backlinks | Builds a backlink profile from free sources |
| /seo google | Pulls GSC, PageSpeed and CrUX data |
Four tasks have evidence behind them:
- Run a full multi-agent site audit with a scored report, per client or monthly (#177, #272, #328).
- Validate schema and generate JSON-LD, per page or template (#143, #187, #188).
- Report Search Console, PageSpeed and CrUX data weekly, with GSC access (#130, #173, #122).
- Build a backlink profile monthly from Moz, Bing and Common Crawl (#100, #153, #312).
GEO and local SEO checks are claimed-only. A run of the local audit was reported (#252), but nobody assessed its results. The 0 to 100 health score uses weights published on claude-seo.md: 23, 22, 20, 10, 10, 10 and 5. Those weights are the publisher's.
Test it yourself:
- Install the plugin.
- Run setup and doctor.
- Run /seo page on one URL you know well.
- Check each finding against the raw HTML and GSC.
How does Claude SEO compare with other Claude SEO skills?
Of the Claude SEO skills named above, Claude SEO alone has a full verdict here, at 70/100. InfraNodus and SE Ranking skills need their own MCP servers. Alex Labat's skill and Hawk Academy's 56 skills are unevaluated.
The criterion is our verdict score. Five related SEO and marketing skills are listed for comparison.
| Option | Type | Needs | Our verdict |
|---|---|---|---|
| Claude SEO (AgriciDaniel) | Full SEO plugin | Python, Chromium, optional keys | Recommended with caveats, 70 |
| Marketing Skills (Corey Haines) | Bundle of 50 marketing skills | Node.js for the npx install | Not rated here |
| SEO Audit (Corey Haines) | Markdown-only audit checklist | A web fetch or browser tool | Not rated here |
| SEOmator SEO Audit | Skill that drives the SEOmator CLI | seomator CLI and Node.js 20.3+ | Not rated here |
| GEO Optimizer (Auriti Labs) | Skill that drives the geo CLI | Python 3.9+ and the geo CLI | Not rated here |
| GEO SEO Claude (zubair-trabzada) | Bundle of 16 GEO and SEO skills | Python 3.10+ and Playwright Chromium | Not rated here |
| InfraNodus SEO skill | Content-gap skill | InfraNodus MCP and account | Not evaluated |
| SE Ranking SEO skills | 7 skills | SE Ranking MCP and paid plan | Not evaluated |
| Claude SEO Skill (Alex Labat) | Single skill | ZIP upload | Not evaluated |
| Hawk Academy SEO skills | 56 skills | None stated | Not evaluated |
Claude SEO covers the widest surface, with technical, schema, local, backlink and Google API work in one plugin. The two single-skill audits carry a lighter runtime. InfraNodus and SE Ranking route data through their own servers; read skills versus MCP servers for that trade-off.
Where is the Claude SEO skill on GitHub?
The Claude SEO skill is on GitHub at AgriciDaniel/claude-seo under the MIT licence. We reviewed v2.4.1, commit ff87fce. The current release is v2.4.2, and the repository has 19,460 stars and 28 contributors.
- Repository: github.com/AgriciDaniel/claude-seo, created on 2026-02-07.
- Licence: MIT.
- Reviewed: v2.4.1, commit ff87fce, released 2026-09-29 and evaluated 2026-09-30.
- Current: v2.4.2, released 2026-10-04.
- Stars: 18,030 at review and 19,460 on 2026-10-21.
- Forks: 2,651 at review.
- Contributors: 28, with 298 commits from AgriciDaniel.
- Releases: 28 in 12 months, and 8 in the 90 days before review.
- Installs: 147.3K across sub-skills on skills.sh at review, with no audits listed there.
- Open issues: #322, #321, #330 and #319.
Version labels differ outside GitHub. claude-seo.md still shows v2.4.1 as "Latest", and agenticskills.io shows v1.7.0. GitHub shows v2.4.2, dated 2026-10-04.
Common questions
Can Claude do SEO?
Yes, with skills or tools that fetch real data. Claude SEO adds scripted audits and Google API checks to Claude Code.
Is Claude good for SEO writing?
This plugin audits sites; it is not a writing skill. For content work, see our writing skills for SEO content.
Is the Claude SEO skill free?
Yes, it is free under the MIT licence. DataForSEO, Ahrefs, Firecrawl and Google Cloud NLP calls bill separately.
Is Claude SEO a plugin or a skill?
It is a Claude Code plugin bundling 35 SKILL.md files, 19 agents and a hook. The skills.sh listing shows an npx install that skips the Python runtime.
Does Claude SEO work on Windows?
Yes, with recurring path, encoding and process bugs (#186, #252). Run /seo doctor after setup.
Is claudeseoskill.com the same plugin?
No. It is Alex Labat's separate Claude SEO Skill, which we have not evaluated.
Who makes Claude SEO?
Daniel Agrici, who publishes as AgriciDaniel on GitHub, makes it. Anthropic does not.
