· 6 min read · By Mehul Prajapati
We Open Sourced an AI Shopify Store Audit: 8 Modules, findings only, no scores
Shopify Cowork audits a storefront across 8 modules using public data only, then reports findings instead of scores. Here is what we built, and why the score had to go.
An AI Shopify store audit does not need store access, and it does not need a score. Shopify Cowork, our open source plugin for Claude Code and Codex, audits any storefront across 8 modules using only public data: Trust, CRO, Page Speed, Technical SEO, Product-Page SEO, Structured Data, AEO, and GEO. It reports concrete findings, and a second skill can implement the fixes with your explicit approval.
We run eagerminds, an AI product studio. We build AI products, ours and yours, and Shopify Cowork is one of ours: two agent skills, one bundled MCP server, and 13+ stars on GitHub. This post is the builder's account: what the audit checks, how to install it, and why we deleted the score.
Why did we build our own Shopify SEO audit?
Because we needed one every week. Before we contact a store owner, we audit their store, and the tools we tried kept handing back the same thing: a colored dial, a number out of 100, and a page of advice that could apply to any store on the internet.
A score like 62/100 reads like a grade and acts like a shrug. It does not tell a merchant that their phone mount page never says which phones fit, or that the return policy is unreachable from the product page. Those are the findings that make an owner actually reply. So we wrote the audit we wanted to receive, packaged it as an agent skill, and open sourced the whole thing.
A score tells a merchant something is wrong. A finding tells them what to fix and on which page.
What do the 8 audit modules check?
store-analyzer runs 8 modules against a live storefront using nothing but public data: no API keys, no admin access, nothing injected into the store. The modules are Trust, CRO, Page Speed, Technical SEO, Product-Page SEO, Structured Data, AEO (answer engine optimization), and GEO (AI visibility).
- Trust: can a first-time buyer verify who they are buying from and how to reach them.
- CRO: friction between landing on a product page and finishing checkout.
- Page Speed: how the storefront actually loads for a visitor.
- Technical SEO: the sitewide crawl and indexing basics.
- Product-Page SEO: titles, descriptions, and content on the pages that sell.
- Structured Data: the machine-readable product markup search engines parse.
- AEO: whether an answer engine can quote the store's pages directly.
- GEO: whether AI assistants can find and cite the store at all.
The last two modules exist because buying behavior moved. People now ask ChatGPT and Perplexity what to buy, and a store those assistants cannot read loses buyers it never sees. Most audit tools still do not look there.
Why findings only, no scores?
Because a score compresses everything useful out of the report. Two stores can both score 62 for completely different reasons, and neither owner learns what to do next. Findings keep the specifics: which page, what is missing, and what a buyer experiences because of it.
We also write findings in buyer language, not tool language. "A buyer cannot tell which phone models fit this mount" gets fixed. "Meta description missing" gets ignored.
There is a second reason: no scores means no false authority. A single number implies a precision the underlying checks cannot honestly claim. A finding can be verified by anyone. Open the page, look, confirm.
How do I install and run it?
Two commands inside Claude Code:
/plugin marketplace add prajapatimehul/shopify-cowork
/plugin install shopify-cowork@shopify-cowork-marketplaceThen point the agent at any storefront:
> audit examplestore.comstore-analyzer does the rest and returns a findings report. The skills are standard skill files, so they also install standalone in any skill-compatible agent. The plugin itself works with both Claude Code and Codex.
Is it safe to let an AI agent write to my Shopify store?
The audit itself never touches your store. It reads public pages only. Writing is the job of the second skill, store-fixer, and it is built to be paranoid: it implements fixes through the Shopify Admin GraphQL API, asks for explicit approval before every write, and saves a rollback manifest for every change it makes.
Bundled tooling
The plugin ships with the official Shopify Dev MCP server (@shopify/dev-mcp) for live docs lookup and GraphQL and theme validation. An agent that guesses API shapes from training data breaks stores. An agent that validates against current Shopify docs first does not.
How do we use it ourselves?
The studio runs this exact skill on real Shopify stores before every piece of outreach we send. We audit a store, read the report the way its owner would, and only then write to them. Some findings survive that reading and some get cut, and the cuts go back into the skill. Open sourcing it means the version on GitHub is the version we depend on, not a demo. The rest of our build notes live on the blog.
FAQ
Does the AI Shopify store audit need access to my store?
No. The store-analyzer skill uses only public data: the pages, markup, and metadata any visitor or crawler can already see. Only the separate store-fixer skill touches the Shopify Admin GraphQL API, and it asks for explicit approval before every write it makes.
Why does the audit report findings instead of scores?
A score compresses dozens of checks into one number a merchant cannot act on. A finding names the exact problem on an exact page, so you know what to fix and can verify the fix yourself. Shopify Cowork reports findings only, across all 8 modules.
Which AI agents does Shopify Cowork work with?
It ships as a plugin for Claude Code and also works with Codex. The two skills, store-analyzer and store-fixer, install standalone in any skill-compatible agent, so you are not locked into one tool. The plugin also bundles the official Shopify Dev MCP server.
Can I roll back changes made by store-fixer?
Yes. store-fixer saves a rollback manifest for every change it applies through the Shopify Admin GraphQL API. Combined with the explicit approval it requests before each write, you always know what changed and have a recorded path back to the previous state.
Built in the open at github.com/prajapatimehul/shopify-cowork.