Claude SEO
Open-source SEO audit plugin for Claude Code — parallel specialist agents that converge into one prioritized action plan
Claude SEO is an open-source plugin that turns Claude Code into an SEO auditor: 25 sub-skills and 18 sub-agents covering technical SEO, content quality, schema, AI-search citability, local and international SEO, dispatched in parallel and synthesized into a single prioritized action plan. It is MIT-licensed, written in Python, and runs locally — no third-party API calls by default, with optional Google credentials for real field data. It will tell you in minutes what is wrong with a site. What it cannot do is the months of publishing and fixing that follow.
What Claude SEO is
Claude SEO (github.com/AgriciDaniel/claude-seo) is an MIT-licensed plugin for Claude Code, written in Python, carrying 13,736 stars as of 9 August 2026. You add it as a Claude Code plugin or clone and run its installer, run `/seo setup` once to build an isolated Python environment plus Playwright Chromium, then drive it with commands: `/seo audit <url>` for a whole site, `/seo page <url>` for one page, plus `/seo technical`, `/seo content`, `/seo schema` and `/seo geo`. A full audit fans out to specialist sub-agents in parallel — up to 15 at a time — and converges into one ranked action plan. There is a companion port for Codex if that is your harness.
What lifts it above a checklist runner is the shape of the output. Every recommendation is required to carry the first-principle observation it rests on, its dependency on other recommendations, an explicit "how would we know this failed?" check, and a leading indicator to watch. The project is also unusually candid about its own boundaries: pages that hydrate on scroll or after a click produce noisy findings, and with no Google credentials attached, Core Web Vitals are lab estimates rather than field data and indexation is inferred rather than confirmed. A four-tier credential system layers in PageSpeed and CrUX, then Search Console, then GA4, then Keyword Planner volume as you supply keys — and credentials stay in a local config directory rather than the repo.
Where it fits a founder's growth stack
For a small team the value is not that it knows things you could not have looked up. It is that it looks at everything, in one pass, and does not get bored on page 40. Three places that pays:
- Before you write anything An audit of a young site tells you whether the pages you already have are indexable, structured and quotable — cheaper to learn now than after you have published on top of a broken foundation. See how do I do SEO for a brand new website.
- When your page count outgrows your attention Templated pages break in templated ways, so one bad pattern is a hundred bad pages at once. That is the programmatic SEO failure mode, and a repeatable audit is the only practical guard against it.
- For AI-search readiness The `/seo geo` command scores passage citability and question-shaped structure — the same levers as generative engine optimization. Its reference material takes explicit, evidence-cited positions on contested tactics instead of repeating them, which is rarer than it should be in this corner of the field.
What an audit cannot do for you
A prioritized action plan is a list of work, not the work. The honest split looks like this:
| What the audit hands you | What still has to happen |
|---|---|
| A ranked list of technical fixes | Someone shipping them into the codebase, in that order |
| Content-quality and E-E-A-T findings | Pages rewritten, sourced, and published on a schedule |
| Schema detected, validated, generated | Markup deployed and re-validated after the next release |
| Citability scores for AI answers | The months of publishing that make you worth citing at all |
We have a stake in this one. AgentCeres — the AI Growth Officer at agentceres.com — builds a managed AI growth team on the same Agent Skills standard this plugin uses, and the SEO pages on this site are authored by an agent routine, so two things about a 25-skill catalogue are worth passing on. First, a skill catalogue is not free: every skill's name and description has to be described to the model inside the same context window as the work, and past a certain size ours stopped fitting — the runtime strips the descriptions and lists bare names, at which point the agent is guessing from a slug what a skill does. More skills is not linearly more capability. Second, depth in a skill's reference files does not move the output on its own. We added deep reference documents to our craft skills and saw nothing change until we also wrote the read ritual — the instruction that says when to open them. Claude SEO's answer to the first problem is a good one worth copying: dispatch sub-agents in parallel so each gets its own context window rather than sharing one crowded prompt.
FAQ
- Do I still need a paid SEO tool alongside it?
- It depends on what you are missing. Out of the box the plugin makes no third-party API calls, so it analyses your own pages rather than competitors' backlinks or keyword volumes. Adding Google credentials unlocks Search Console, GA4 and Keyword Planner data, and MCP extensions connect commercial data providers if you already pay for one. If your question is "is my site technically sound and citable?" the free tier answers it. If your question is "what are my competitors ranking for?" you still need a data subscription.
- Does it work outside Claude Code?
- The plugin itself targets Claude Code and installs through its plugin marketplace or a shell script. The project maintains a separate Codex-first port with the same workflow surface for people running that harness. There is no browser app or hosted version — this is a local tool that runs on your machine against URLs you point it at.
- Is it safe to run against a production site?
- It fetches the URLs you give it, the way any crawler would, and stores any credentials you add in a local config directory with restrictive permissions rather than in the repo. The ordinary caution applies to installation rather than crawling: it is a local tool that executes code, and the project itself recommends cloning and reading the installer instead of piping a remote script into a shell. Read it first, as you would with any tool that earns filesystem access.
- Will an audit make my site rank?
- No — it produces diagnosis and priority, and rankings follow the shipping. That is the honest limit of every audit tool, paid or free: the report is the cheap half. Budget your energy accordingly, because the value shows up when the top three items get fixed, not when the report is generated.
You built it. Now grow it.
AgentCeres is a managed AI marketing team — specialists draft the SEO, social, and outreach that fill your links, you approve what ships. 14-day free trial, from $39/month.