Faro AI Signals·July 20, 2026

MCP Goes Stateless, llms.txt Hype Cracks, Citations Favor Lists

The Model Context Protocol's biggest revision yet strips out session state and adds routable headers — reshaping how AI agents connect to business services. Meanwhile, two independent datasets puncture the idea that publishing llms.txt alone drives AI citation.

ShareShare

This Week's Signals

MCP's July 28 Release Candidate Strips Sessions and Adds Gateway-Ready Routing Headers

Source: Model Context Protocol Blog ↗

Why it matters for your score

The protocol is now fully stateless — no initialize handshake, no sticky sessions — which means businesses can put standard load balancers in front of MCP servers without custom infrastructure. New Mcp-Method and Mcp-Name headers let gateways route by operation without reading request bodies, and a new caching model (ttlMs and cacheScope) means repeated AI-agent reads of your data no longer require a fresh server hit every time. If your team is building or buying MCP server tooling, the 10-week SDK update window starts now.

Shopify Auto-Pushed llms.txt to 78% of Stores With No Merchant Notice — Platform Decisions Now Dominate Adoption Numbers

Source: Casey Burridge (via HTTP Archive BigQuery) ↗

Why it matters for your score

The apparent 5x growth in llms.txt adoption over 12 months is substantially a Shopify infrastructure decision, not organic business action — 78% of Shopify stores got the file silently in late April 2026. For non-Shopify businesses, adoption requires deliberate steps; WordPress sits at 8.7% and nearly 40% of all existing files are auto-generated stubs that likely provide no citation benefit. Publishing a file is necessary groundwork, but it is not sufficient on its own.

63% of LLM Citations Go to Listicle Pages and 44% Come From a Document's First Third

Source: Digital Applied (citing Evertune / Wix research) ↗

Why it matters for your score

Across nearly 400 million citations from ChatGPT, Gemini, Perplexity, and others, the structure of your content predicts citation more than the presence of any metadata file. Leading with your key claims, formatting content as numbered Top-N lists, and publishing original data all carry measurable citation weight — original research pages earn more than three times their proportional share of citations. Businesses should audit whether their highest-value pages front-load their strongest claims and use structured comparison formats.

What to do this week

1
Review whether your MCP server implementation relies on session state or sticky-session load balancing — the July 28 spec makes both obsolete. Use the /tools/ai-readiness-scan to identify integration points that need updating before the 10-week SDK window closes.
2
If you are not on Shopify, do not assume llms.txt is handled for you. Use the /tools/llms-txt tool to generate a properly structured file — and verify it is not an empty stub — since the SE Ranking data shows 40% of existing files deliver no measurable benefit.
3
Restructure your top commercial pages to lead with the key claim in the first third of the document and include at least one numbered comparison list. Use the /tools/okf-generator to add structured data that reinforces the hierarchy AI models use when extracting citations.

Sources

← All editions Scan your site now →