Anthropic Expands MCP to Web; LLMs.txt Hits 50K Domains
Two structural shifts this week: Anthropic's MCP specification crossed into broad web deployment with new third-party hosting support, and llms.txt adoption surpassed 50,000 indexed domains — signaling that machine-readable business identity is moving from early-adopter experiment to baseline expectation.
This Week's Signals
Anthropic Opens MCP Server Hosting to Third Parties, Accelerating Agent-Readable Business Endpoints
Why it matters for your score
Anthropic's Model Context Protocol now supports third-party hosted MCP servers, meaning businesses can expose structured, agent-queryable endpoints without running their own infrastructure. For a local service or SaaS company, this is the difference between being invisible to Claude-powered agents and being actively callable — agents can pull your hours, pricing, and availability in real time rather than scraping a static page. Businesses that publish an MCP-compatible endpoint this month will be indexed before the majority of competitors have heard of the spec.
LLMs.txt Adoption Crosses 50,000 Domains as Perplexity Confirms Active Parsing
Why it matters for your score
Perplexity has confirmed its crawler actively parses llms.txt files to inform answer sourcing, joining a growing list of AI systems that treat the file as a trust and relevance signal. With 50,000 domains now publishing one, the file has cleared the threshold where absence becomes a visible gap rather than a neutral default. If your llms.txt is missing or malformed, you are being ranked below competitors in AI-surfaced recommendations — not hypothetically, but in live query results today.
Google's Structured Data Documentation Updated to Reference AI Agent Parsing Behavior
Why it matters for your score
Google quietly updated its structured data documentation to explicitly note that JSON-LD markup is consumed by AI Overviews and agent-mode queries — not just traditional search ranking. This confirms that Schema.org fields like 'hasOfferCatalog', 'priceRange', and 'areaServed' are being read by the AI layer, not just the crawl index. Businesses with thin or missing JSON-LD are losing citation surface area in AI Overviews, which now appear on a significant share of commercial-intent queries.
What to do this week
- 1
Audit your llms.txt file this week against Perplexity's confirmed parsing behavior — specifically, ensure your 'services', 'pricing', and 'coverage-area' fields are populated with exact, machine-readable values rather than prose descriptions. Use Faro's /tools/llms-txt generator to rebuild from a structured template if your current file was created manually or more than 90 days ago.
- 2
Run your domain through Faro's /tools/ai-readiness-scan and filter results by MCP exposure score — if you score below 40, prioritize publishing a basic MCP-compatible endpoint using Anthropic's new third-party hosting pathway before Q3 indexing cycles complete. Even a read-only endpoint covering hours, pricing, and service area is sufficient to become agent-callable.
- 3
Update your JSON-LD schema to include 'hasOfferCatalog' and 'priceRange' markup in response to Google's AI agent parsing confirmation — use Faro's /tools/ai-schema to validate that these fields render correctly and are not being blocked by your robots.txt for GPTBot, ClaudeBot, or PerplexityBot. Blocking any of the three major AI crawlers from your schema endpoints is now a direct revenue risk.