The AgenticCart MCP server lets any AI assistant that speaks the Model Context Protocol run AgenticCart's store audits directly. Ask Claude, Cursor or your own agent whether a store is ready for AI shopping assistants, and it checks the live site and answers with a score and specific findings. It works on any store on the public web and needs no account to try: without credentials it returns the score, how many issues there are, and the biggest one. Every AgenticCart plan unlocks the complete findings.
What the MCP server does
It exposes the same five checks that power the free tools, as callable tools instead of web forms. The difference matters in practice: a web form gives you a report you then have to act on by hand, while a tool result lands in the same conversation as everything else your assistant can do. It can read the findings, explain them in context, and write the fix.
The five tools
- check_store_readiness - audits a whole store from its homepage URL. Returns a 0-100 score covering AI crawler access in robots.txt, structured data, sitemap, llms.txt and HTTPS.
- check_product - audits a single product page for how well AI shopping assistants can read and recommend it: Product and Offer schema, price and availability markup, images, description quality, GTIN and brand.
- check_acp_feed - grades a product feed against the Agentic Commerce Protocol and the OpenAI product feed specification, which govern whether a catalog can be sold inside ChatGPT.
- check_merchant_feed - validates a product feed against the Google Merchant Center product specification: required attributes, price formatting, availability values and GTIN checksums.
- check_llms_txt - reads a site's /llms.txt and checks it against the llmstxt.org format: title, summary, link sections, whether the listed links answer, and whether the server sends the file rather than an HTML page.
Connect it
The server is a remote HTTP endpoint, so there is nothing to install and nothing to keep running on your machine.
Endpoint: https://agenticcart.ai/mcp
For an overview with copyable setup snippets, example prompts and the full tool reference, see the MCP server page.
In Claude Code, add it with one command:
claude mcp add --transport http agenticcart https://agenticcart.ai/mcp
In a client that is configured through a JSON file, such as Claude Desktop or Cursor, add the server to your MCP configuration:
{
"mcpServers": {
"agenticcart": {
"url": "https://agenticcart.ai/mcp"
}
}
}
Then ask your assistant something like "check whether example.com is ready for AI shopping assistants" and it will call the right tool on its own.
Free tier and full results
Two tiers, and the difference is the remediation detail rather than the diagnosis. All five tools work on both, and every issue is named on both.
- Free, no account - the readiness score and grade, and every finding named and grouped by severity, so the diagnosis is complete. One finding also carries its remediation detail, as a sample. Ten checks per hour, keyed by IP address.
- Any AgenticCart plan - the remediation detail on every finding, so your assistant can write the fix, plus the raw structured result for it to reason over. 300 checks per hour, keyed by your account rather than your address, so a team behind one shared connection does not share one budget.
To unlock the full results, take the licence key from Dashboard > API Key and send it as a Bearer token:
claude mcp add --transport http agenticcart https://agenticcart.ai/mcp \
--header "Authorization: Bearer ac_live_your_key_here"
There is no separate MCP subscription. If the key is missing, wrong, or the subscription is no longer active, the server returns free-tier results and states the reason rather than failing.
Both rate limits are separate from the ones on the web tools, so heavy use of the MCP never locks anyone out of the checkers on the website, and the other way round.
What it does not do
The MCP server is read-only. It fetches publicly reachable URLs and analyses them. It cannot change a store, cannot read anything behind a login, and has no access to any AgenticCart account, catalog or conversation data. Requests to private or internal network addresses are rejected.
When to use the MCP instead of the web tools
Use the web tools when you want a shareable report for one store, with the full visual breakdown. Use the MCP when you are already working in an AI assistant: reviewing several stores in a row, auditing a client's catalog while writing the fixes, or wiring store checks into your own agent. Both run exactly the same analysis, so the scores match.
Making a store pass
Most low scores come down to the same handful of causes: AI crawlers blocked in robots.txt, missing Product and Offer structured data, no llms.txt, and feeds without GTIN or MPN identifiers. The first three are site-level fixes described in the agentic storefront. The identifier problem is a catalog-data problem and usually needs fixing at the source, in WooCommerce or Shopify.
If the goal is not just to be readable by AI assistants but to sell through one, see how AgenticCart works: connecting a store turns the catalog into a hosted AI storefront on your own domain, rather than hoping external crawlers read it correctly.
See your own store in AgenticCart
Enter your shop’s address. The scan shows what ChatGPT, Claude and Gemini can read today and prepares your products for them, in about a minute.
Or start the free 7-day trial right away.