A store with 40 products can be run by hand. A store with 4,000 cannot. Somewhere between those numbers, the work that matters most for organic traffic (titles, meta descriptions, collection copy, alt text, feed attributes) stops getting done, because nobody has the hours.
Claude Code can do a lot of that work, but only if it knows how your catalog is shaped and how Shopify’s APIs behave today. Out of the box it guesses. Skills and MCP servers replace the guessing with instructions and live documentation.
This guide covers the tools that do e-commerce work: Shopify’s own AI Toolkit and Dev MCP server, store-connected plugins, catalog SEO skills, and paid packs. Every GitHub number was checked through the GitHub API on 2 November 2026. This is a research-based roundup. We read the READMEs, docs and pricing pages; we did not run each tool against a live store, and we say so wherever that matters.
This page is about the store and catalog side (Shopify APIs, product and collection SEO, feeds). For paid social, Klaviyo flows and UGC, see Claude Code skills for D2C brands.
If skills are new to you, read what Claude Code skills are first.
The short answer: what should a Shopify store install first?
Install Shopify’s official AI Toolkit plugin for correct API calls, add StoreHand if you want a read-only daily briefing and an SEO health check, and use claude-seo or marketingskills for the content and audit work. All four are free. Paid packs are optional and none is Shopify-specific.
Why that order? Because most damage in AI-assisted store work comes from bad API calls, not bad prose. A wrong GraphQL mutation can overwrite 300 product descriptions. Shopify’s toolkit validates queries against the current schema before they run. Once that floor exists, the content skills can do their job.
The full shortlist at a glance
| # | Tool | Type | Free or paid | Best for | GitHub stars (2 Nov 2026) | Last push |
|---|---|---|---|---|---|---|
| 1 | Shopify AI Toolkit | Plugin + skills (official) | Free (MIT) | Docs, validation, store ops via CLI | ~623 | 27 Oct 2026 |
| 2 | Shopify Dev MCP server | MCP server (official) | Free | Docs and schema lookup | npm package, no public repo found | n/a |
| 3 | StoreHand | Plugin | Free (Apache-2.0) | Daily briefings, health check, listing copy | 2 | 5 Aug 2026 |
| 4 | shopify-mcp (GeLi2001) | Community MCP server | Free (MIT) | Admin API CRUD on products, orders, metafields | ~239 | 5 Apr 2026 |
| 5 | claudekit-skills Shopify skill | Skill | Free (no licence file) | Theme and app development | ~2.2k | 3 Apr 2026 |
| 6 | claude-seo | Skill pack | Free (MIT) | Technical SEO, schema, GEO audits | ~25.6k | 1 Nov 2026 |
| 7 | marketingskills | Skill pack | Free (MIT) | Copy, CRO, email | ~57.9k | 31 Oct 2026 |
| 8 | claude-skills (alirezarezvani) | Skill library | Free (MIT) | Broad marketing bundle | ~29.1k | 31 Oct 2026 |
| 9 | Anthropic Marketing plugin | Plugin (official) | Free | Campaign plans, brand review | see repo | see repo |
| 10 | Inflectual Master Marketing Skills Bundle | Paid pack | $167 one-time | Large general marketing library | n/a | n/a |
| 11 | Contra Ultimate Claude SEO & AEO Skills Pack / SuperSkills | Paid packs | $39 (shown vs $79) / $50 | SEO and AEO prompts in skill format | n/a | n/a |
Star counts move daily. A two-star repo can be well built and a 50k-star repo can be wrong for your store, so read the files before you install.
How we picked (and what we left out)
A tool made the list if it does real store work, publishes readable source or official docs, and shows recent activity. We checked each repo through the GitHub API for stars, licence, last push and archived status, and read each README for install steps and permissions.
Three filters mattered most:
- Permissions disclosed. Store-connected tools touch orders, customers and money. If a README does not say which Admin API scopes it needs, it was not listed.
- Shopify accuracy. Shopify’s APIs change on a quarterly version cycle. A skill frozen on an old API version will write mutations that fail or, worse, half succeed. We flag the ones that admit their snapshot date.
- E-commerce over general dev. Plenty of “Shopify skills” are app-development scaffolds. Useful for agencies, less so for a merchant who wants better collection pages.
What we left out
- Four tiny Shopify skill repos found by GitHub search (zero stars, no README detail on scopes). They may be fine; we could not judge them.
- Closed-source “AI for Shopify” SaaS apps. They are not Claude Code skills.
- Any pack that relies on fake countdown timers to sell. We do not recommend vendors who manufacture urgency.
Shopify’s official tools: AI Toolkit and Dev MCP server
Yes, Shopify ships official Claude Code tooling. The Shopify Dev MCP docs list two routes: the AI Toolkit plugin (recommended) and the standalone Dev MCP server. The Dev MCP runs locally and, per the docs, does not require authentication.
Install commands (from shopify.dev, 2 November 2026)
Plugin, recommended by Shopify:
claude plugin install shopify-ai-toolkit@claude-plugins-officialDev MCP server only:
claude mcp add --transport stdio shopify-dev-mcp -- npx -y @shopify/dev-mcp@latestRestart Claude Code after either. Some third-party guides show an older two-step /plugin marketplace add Shopify/shopify-ai-toolkit flow; use whatever Shopify’s page shows on the day you install.
What it does
The Shopify-AI-Toolkit README describes three capabilities:
- Search Shopify developer docs and API schemas
- Validate GraphQL, Liquid and UI-extension code
- Manage a store “through the CLI’s store execute capabilities”
That third line is the one to read twice. It means the agent can act on a store you have authenticated with Shopify CLI. Treat it like handing a contractor admin access.
Telemetry: read this before installing
The README states telemetry is on by default. Skill scripts send usage events to shopify.dev/mcp/usage, which can include search queries, validation results and validated code. On Claude Code, a hook can attach your most recent prompt, truncated to 2,000 characters. The recommended opt-out is to create the file ~/.config/shopify-ai-toolkit/opt-out. Environment variables (OPT_OUT_INSTRUMENTATION=true or DO_NOT_TRACK=1) also work, but the README warns some hosts do not pass them through.
If your prompts include supplier pricing or customer details, create that opt-out file first.
What happened to Storefront MCP
Shopify’s Storefront MCP page now says its catalog and cart tools have been removed in favour of the Universal Commerce Protocol (UCP) at https://{shop}/api/ucp/mcp, with Storefront Catalog MCP (search_catalog, lookup_catalog, get_product), Cart MCP, Checkout MCP and Ask MCP. Many 2026 blog posts still describe the old /api/mcp endpoint. These servers serve shopping agents browsing your store, not you editing it, but they matter for one reason: AI shopping agents now read your catalog data directly. Clean titles and attributes are no longer only an SEO concern.
Takeaway: install the official toolkit for correctness, opt out of telemetry if needed, and keep its store-execute ability pointed at a development store until you trust your workflow.
Store-connected plugins and MCP servers
For a merchant, StoreHand is the most store-shaped plugin we found, and GeLi2001’s shopify-mcp is the most complete Admin API server. They sit at opposite ends of the risk scale.
StoreHand: read-only by default
StoreHand is a Claude Code plugin billed as “an extra pair of hands for your store.” It needs Shopify CLI 4.5 or newer. Install:
/plugin marketplace add storehand/storehandThen run /storehand:storehand-setup. Setup requests five read-only scopes. Its skills, per the README:
| Skill | What it does | Writes to store? |
|---|---|---|
| daily-store-briefing | Orders, failed payments, refunds, low stock | No |
| store-health-check | Sold-out active products, discount issues, broken links, missing SEO metadata | No |
| product-listing-writer | Drafts titles, descriptions, SEO fields and alt text into a file you edit | Only approved changes, with optional write_products and write_files |
| weekly-store-report | Week-over-week revenue, orders, AOV, sessions, conversion | No |
The repo had 2 stars when we checked. That is a reason to read the code, not a reason to skip it. The draft-to-file, approve-then-apply pattern is exactly what a store should want.
shopify-mcp: full Admin API access
GeLi2001/shopify-mcp exposes 31 tools across products, variants, customers, orders, draft orders, metafields, inventory and tags. That includes refund-create, order-cancel (the README notes cancellation is irreversible) and delete-product. It authenticates with Dev Dashboard client credentials (required for new apps since January 2026, per the README) or a legacy shpat_ token.
claude mcp add shopify -- npx shopify-mcp --clientId YOUR_CLIENT_ID --clientSecret YOUR_CLIENT_SECRET --domain your-store.myshopify.comGive the app only the scopes a task needs. For catalog SEO, that is read_products and write_products. Do not grant order or customer write scopes to a session whose job is meta descriptions.
claudekit-skills Shopify skill
mrgoonie/claudekit-skills includes a Shopify development skill covering apps, extensions, themes and GraphQL/REST. A directory listing says it reflects the platform as of January 2025, and the repo has no licence file. Use it for theme scaffolding ideas, and let the official toolkit validate whatever it produces.
Catalog SEO skills: titles, meta, alt text and schema
Catalog SEO is where Claude Code pays back fastest, because the work is repetitive, rule-based and huge. A good catalog skill reads an export, applies your rules per row, and writes a reviewable file. It never pushes straight to the store.
The general SEO packs (claude-seo, marketingskills) do not know Shopify’s quirks, so you add those in a short project skill. What the skill should encode:
- Title tag pattern. For example:
{Product} {Key attribute} | {Brand}, capped near 60 characters. - Meta description rules. Lead with the use case, include one attribute that differs by variant, no repeated boilerplate across a collection.
- Alt text. Describe what the image shows, not the keyword list.
- Duplicate checks. Flag rows whose description shares long runs of text with another row.
- Schema check. Most Shopify themes already output Product JSON-LD. The skill should check, not duplicate. Google’s Product structured data docs list the properties.
Example prompt
Read
products_export.csv. For every active product missing a SEO title or SEO description, draft both using the rules incatalog-seo/SKILL.md. Write results toseo_drafts.csvwith columns Handle, Current title, New title, New description, Character counts, Duplicate flag. Do not call any Shopify write tool.
What the output should contain
- One row per product handle, unchanged handles
- Character counts for each field so you can sort and spot overruns
- A flag column for rows the skill was unsure about (no attributes, no images)
- No invented specs: if the CSV lacks a material or size, the description should not claim one
Descriptions themselves have their own playbook in Claude Code skills for product descriptions. For structured data depth, see skills for schema markup, and for a full crawl, technical SEO skills.
Takeaway: general SEO skill for the method, a 40-line project skill for your store’s rules, file output for review.
Product feeds: Google Merchant Center and beyond
No dedicated, well-maintained Claude Code skill for product feeds turned up in our search, so this is a build-your-own job. That is less work than it sounds, because a feed is a spreadsheet with a published spec.
Google’s Merchant Center product data specification defines the attributes: id, title, description, link, image_link, availability, price, brand, gtin, mpn, condition and category fields among others. A feed skill should:
- Map your Shopify export columns to spec attributes
- Flag rows missing required attributes for their category
- Rewrite feed titles with attribute order that suits the category (brand, product type, size, colour, for apparel; brand, model, key spec, for electronics)
- Never alter
price,gtinoravailability. Those come from the source of truth, not from a language model
Example prompt
Using the Merchant Center spec in
feed-rules.md, auditfeed.tsv. Outputfeed_issues.csvlisting each item id, the missing or invalid attribute, and a suggested fix. Then writefeed_titles.csvwith proposed titles only. Leave price, GTIN and availability untouched.
What the output should contain
- An issues file sorted by count, so the most common error gets fixed first
- Proposed titles in a separate file from factual fields
- A note on any GTIN that fails its check digit, since that is a data problem, not a copy one
Why the strict separation? A model that “improves” a price string or guesses a GTIN creates a feed disapproval, or worse, a mismatch between ad and checkout. Keep the model on wording and keep numbers mechanical.
Feeds also feed paid shopping campaigns. For the ad side, see Claude Code skills for PPC.
Collection pages: the most neglected SEO asset in Shopify
Collection pages usually target the commercial keywords that matter most (“linen shirts for men”, “trail running shoes”), yet most stores leave them as a product grid with no copy. Claude Code can draft the intro copy, FAQs and internal links at scale, provided a skill keeps it honest and specific.
A collection skill should produce, per collection:
- A 60 to 120 word intro above the grid that names the product type and the main buying decision
- An optional below-grid section with buying guidance (sizing, materials, care) built from your actual product attributes
- Two or three FAQs drawn from real customer questions (support tickets, reviews), not invented ones
- Links to sibling and parent collections
- A check for filtered or tagged URLs that create thin duplicates of the same collection
Example prompt
For each collection in
collections.csv, read the products listed under it inproducts_export.csv. Draft an intro (max 120 words) that names the common materials, price range and use cases found in the actual products. Add two FAQs only ifsupport_questions.mdcontains matching questions. Outputcollection_drafts.md, one section per collection handle.
What the output should contain
- Facts drawn from the product data (price range computed, not estimated)
- A marker where the skill had no source for an FAQ, instead of a made-up question
- Suggested internal links using existing collection handles only
Collection architecture is a topic of its own. Topical map skills help plan which collections deserve to exist, and internal linking skills help connect them. Large catalogs with templated category pages should also read programmatic SEO skills.
Reviews: summaries, responses and the legal limits
Claude Code is useful for reading reviews and drafting replies, and dangerous if pointed at writing them. The line is clear in US law.
Useful review workflows
- Theme mining. Export reviews, ask a skill to cluster complaints and praise by product, and produce a table of recurring issues. That feeds product fixes and collection FAQ content.
- Response drafts. Draft replies to negative reviews in your brand voice, for a human to edit and post.
- Copy input. Pull the exact words customers use (“runs small”, “doesn’t pill”) into a phrase bank for descriptions.
Example prompt
Read
reviews.csv. For each product with 10+ reviews, list the top three praised attributes and top three complaints, quoting the review IDs that support each. Do not paraphrase into claims the reviews do not make. Outputreview_themes.md.
The compliance limits
- The FTC’s Trade Regulation Rule on the Use of Consumer Reviews and Testimonials (16 CFR Part 465) prohibits fake reviews, including AI-generated ones attributed to people who did not have the experience, and the suppression of negative reviews. It took effect in October 2024.
- The FTC’s Endorsement Guides FAQ warns that incentives tied to positive reviews can distort them and that fake engagement is deceptive.
- Google’s review snippet guidelines require reviews in markup to be genuine and visible on the page.
So: a skill can summarise, theme and draft replies. It must never generate review text, star ratings or “customer quotes” for publication. Put that rule in the skill’s SKILL.md in plain words. Brand-voice guidance for replies lives in skills for brand voice.
Paid Claude Code skill packs: worth it for a store?
Probably not as a first purchase. None of the paid packs we verified is built for Shopify. They are general marketing or SEO libraries. If the free tools above cover your needs, they cover them.
Where a paid pack can make sense is when you want a large, pre-written prompt library and do not want to assemble one from GitHub.
| Pack | Price (verified) | Shopify-specific? | Notes |
|---|---|---|---|
| Inflectual Master Marketing Skills Bundle | $167 one-time, 167 skills | No | Broad marketing coverage; skill count is the seller’s figure |
| Contra Ultimate Claude SEO & AEO Skills Pack | $39 shown vs $79 | No | SEO and answer-engine focus, applicable to catalog SEO |
| SuperSkills | $50 | No | General skill pack |
| DataForSEO (data, not skills) | Pay-as-you-go, $50 minimum top-up, $1 trial | No | Keyword and SERP data that SEO skills can call |
And Claude itself: a Claude Pro plan is $20 a month ($17 a month billed yearly) and includes Claude Code.
We did not buy or test these packs. Prices were taken from vendor pages and our earlier verified checks. Before buying, ask for a file list and confirm whether you get editable SKILL.md files or a PDF of prompts.
Free vs paid at a glance
| Need | Best free option | Paid option | Verdict |
|---|---|---|---|
| Correct Shopify API calls | Shopify AI Toolkit | None needed | Free wins |
| Store briefings and health checks | StoreHand | None found | Free wins |
| Bulk Admin API edits | shopify-mcp (scoped) | None found | Free, with care |
| Catalog SEO audits | claude-seo | Contra SEO & AEO pack | Free first |
| Product and collection copy | marketingskills + project skill | Inflectual bundle | Free first, paid for volume of templates |
| Feed optimisation | Build your own | None found | Build |
| Keyword data | Google Search Console exports | DataForSEO | Paid if you need volumes |
A store workflow that chains these tools
Run the store in passes: audit, draft to file, review, apply in small batches, measure. Each pass uses a different tool.
- Audit (read-only). StoreHand’s
store-health-checkorclaude-seolists missing SEO fields, sold-out active products and broken links. - Draft catalog fields. Your catalog skill writes
seo_drafts.csvfrom the product export. - Draft collection copy. The collection skill writes
collection_drafts.mdfrom product data and support questions. - Review. A human edits both files. This step is not optional.
- Validate the write. With the AI Toolkit connected, ask Claude to write the bulk-update GraphQL and validate it against the current schema.
- Apply to 20 products first. On a development store or a small batch, then check the storefront.
- Measure. Compare clicks and impressions in Search Console for the edited URLs after a few weeks. We have no benchmark for what change to expect, and nobody honest does without your data.
Install mechanics for each step are in how to install Claude Code skills. If the terminal is the barrier, Claude Code for non-developers covers that.
Safety rules for any store-connected setup
Treat every store-connected agent like a new staff member on day one: limited access, supervised, logged. Several third-party guides report that the AI Toolkit’s store operations offer no rollback in the current release. We could not confirm that on shopify.dev, so assume it is true until you see otherwise.
- Use a development store or duplicated products for the first runs
- Grant the minimum Admin API scopes per task
- Export a product CSV backup before any bulk write
- Keep skills in “write to file” mode until output has passed review several times
- Log every mutation the agent runs, with timestamps
- Opt out of toolkit telemetry if prompts may contain customer or supplier data
- Never let a skill alter price, inventory or GTIN fields from generated text
These rules are cheap. Restoring 2,000 product descriptions from memory is not.
Common mistakes stores make with Claude Code
The biggest mistake is letting the model write facts it was never given. A description that claims “organic cotton” when the product data says nothing about fibre is a returns problem and possibly a consumer-protection one.
Other mistakes we see described repeatedly in guides and forum posts:
- Trusting training data for Shopify APIs. Connect the Dev MCP or toolkit so mutations match the current version.
- One prompt for the whole catalog. Different categories need different title orders and description lengths. Encode category rules in the skill.
- Boilerplate across variants. 200 near-identical descriptions help nobody. Add a duplicate check.
- Skipping collections. Teams polish product pages and leave the higher-intent collection pages empty.
- Ignoring telemetry and scopes. Defaults are rarely the right setting for a store with customer data.
- Buying a pack before trying the free tools. None of the paid packs is Shopify-specific.
When nothing fits, write your own skill. It is a Markdown file with a name and a description: how to build a Claude Code skill.
FAQ
Does Shopify have an official MCP server for Claude Code?
Yes. The Shopify Dev MCP server runs locally, needs no authentication, and helps with docs, API schemas and code validation. Shopify recommends its AI Toolkit plugin, which bundles that ability with skills.
How do I install the Shopify AI Toolkit in Claude Code?
Per shopify.dev on 2 November 2026: claude plugin install shopify-ai-toolkit@claude-plugins-official, then restart Claude Code.
Can the Shopify AI Toolkit change my live store?
Its README says it can manage a store through Shopify CLI’s store execute capabilities. Test on a development store first.
Does the Shopify AI Toolkit collect data?
Yes, telemetry is on by default and can include queries and validated code. Create ~/.config/shopify-ai-toolkit/opt-out to opt out.
What happened to Shopify’s Storefront MCP?
Shopify’s docs say its catalog and cart tools were removed in favour of the Universal Commerce Protocol, served at /api/ucp/mcp.
What is the best free Claude Code skill for Shopify SEO?
For audits, claude-seo. For store-specific checks like missing SEO metadata, StoreHand’s health check. Pair either with a short project skill holding your title and meta rules.
Is there a Claude Code skill for Google Merchant Center feeds?
We found no well-maintained dedicated one. Build a skill around Google’s product data specification and keep price, GTIN and availability untouched.
Can Claude Code write collection page copy?
Yes, if it reads your actual product data. Limit facts to what the export contains and use real customer questions for FAQs.
Can I use Claude to write product reviews?
No. The FTC’s consumer reviews rule bans fake reviews, including AI-generated ones. Use Claude to summarise and respond to real reviews.
Do I need to know how to code?
Not for file-based workflows like catalog SEO drafts. Store-connected MCP servers need comfort with API credentials and scopes.
Is it safe to connect Claude Code to my Shopify Admin API?
It can be, with scoped credentials, backups, a development store and logged mutations. Community servers like shopify-mcp include destructive tools such as refunds and deletes.
Will these skills work with WooCommerce or BigCommerce?
The content and SEO skills work on any CSV export. The Shopify AI Toolkit and Shopify MCP servers are Shopify-only.
How much does it cost to run Claude Code for a store?
Claude Pro is $20 a month ($17 a month yearly) and includes Claude Code. Every tool in the free section costs nothing extra.
Are paid skill packs worth it for e-commerce?
Not as a first step. None we verified is Shopify-specific. Consider one only if you want a large prewritten template library.
Will AI-written product content hurt my rankings?
Google’s guidance targets unhelpful content made mainly for rankings, regardless of how it is produced. Specific, accurate, reviewed copy is the safer path. We have no ranking data of our own on this.
How often should I rerun catalog audits?
Weekly suits busy catalogs, which matches StoreHand’s weekly health check design. Smaller stores can run monthly.
Did you test these tools on a live store?
No. This guide is research-based: we read docs, READMEs and pricing pages and checked GitHub data. We did not run them against a store.
Sources and methodology
Repo stars, licences, last push and archived status came from the GitHub API on 2 November 2026. Install commands came from shopify.dev and each README. Prices came from vendor pages and earlier verified checks. Reddit could not be reached by our research tools, so no Reddit threads are quoted; forum sentiment mentioned above is labelled as unverified.
- Shopify Dev MCP: https://shopify.dev/docs/apps/build/devmcp
- Shopify Storefront MCP (UCP migration): https://shopify.dev/docs/apps/build/storefront-mcp
- Shopify-AI-Toolkit: https://github.com/Shopify/Shopify-AI-Toolkit
- StoreHand: https://github.com/storehand/storehand
- GeLi2001/shopify-mcp: https://github.com/GeLi2001/shopify-mcp
- mrgoonie/claudekit-skills: https://github.com/mrgoonie/claudekit-skills
- AgriciDaniel/claude-seo: https://github.com/AgriciDaniel/claude-seo
- coreyhaines31/marketingskills: https://github.com/coreyhaines31/marketingskills
- alirezarezvani/claude-skills: https://github.com/alirezarezvani/claude-skills
- Anthropic Marketing plugin:
claude plugin marketplace add anthropics/knowledge-work-pluginsthenclaude plugin install marketing@knowledge-work-plugins - Google Merchant Center product data spec: https://support.google.com/merchants/answer/7052112
- Google Product structured data: https://developers.google.com/search/docs/appearance/structured-data/product
- Google review snippet guidelines: https://developers.google.com/search/docs/appearance/structured-data/review-snippet
- FTC consumer reviews rule: https://www.ftc.gov/legal-library/browse/rules/rule-use-consumer-reviews-testimonials
- FTC Endorsement Guides FAQ: https://www.ftc.gov/business-guidance/resources/ftcs-endorsement-guides-what-people-are-asking
Your next step
Install the Shopify AI Toolkit, create the telemetry opt-out file, and export your products CSV. Then ask Claude Code for one thing: a list of every active product missing an SEO title or description, written to a file. That single read-only run shows you the size of the job and how well the tooling fits your store, before anything touches a live listing.
