Anthropic Ships MCP Registry as Claude Agent Routing Goes Live
Anthropic's official MCP Registry entered public availability this week, giving Claude agents a structured directory to discover and invoke business-exposed tools — the first time a major AI lab has operationalized a crawlable, permission-scoped index of agent-facing endpoints. For businesses, this is the moment MCP exposure shifts from experimental to competitively necessary.
This Week's Signals
Anthropic Launches Public MCP Registry, Enabling Claude to Route Tasks to Listed Businesses
Why it matters for your score
The MCP Registry functions like a DNS layer for AI agents: Claude can now query it to find which businesses have exposed compliant tool endpoints, then invoke them directly during agentic tasks. A business not listed in the Registry — or whose MCP server returns malformed schemas — will be invisible to Claude when it's acting autonomously on a user's behalf. Run Faro's /tools/ai-readiness-scan now to check whether your MCP endpoint is registry-compliant before your competitors are listed and you are not.
Perplexity Confirms Citation Weighting Shift: Structured Product and Pricing Data Now Prioritized
Why it matters for your score
Perplexity's engineering team confirmed this week that its answer engine now scores pages with valid JSON-LD Product, Offer, and FAQPage schema meaningfully higher when constructing shopping and service recommendation answers. Unstructured pricing pages — plain text rates buried in paragraphs — are being deprioritized in favor of machine-readable Offer blocks with priceValidUntil and priceCurrency fields. If your pricing page isn't marked up, you are actively losing recommendation slots to competitors who have done the 20-minute schema implementation. Use Faro's /tools/ai-schema to generate compliant JSON-LD for your service or product pages.
GPTBot Crawler Documentation Updated: OpenAI Adds Selective Section Blocking via robots.txt Directives
Why it matters for your score
OpenAI quietly updated GPTBot's official documentation to support path-level allow/disallow directives, meaning businesses can now expose specific high-value pages — pricing, service descriptions, llms.txt — to GPTBot while blocking thin or duplicate content from crawling. Previously, it was all-or-nothing. This matters because curating what GPTBot indexes directly shapes what surfaces in ChatGPT's browse-and-recommend responses. Audit and update your robots.txt using Faro's /tools/robots-txt to ensure GPTBot is reading your best content and not your boilerplate.
What to do this week
- 1
Submit your MCP endpoint to Anthropic's public MCP Registry this week — not next month. Use Faro's /tools/ai-readiness-scan to validate that your server returns a compliant tool manifest with correct authentication scopes before submission; a malformed registration is worse than no registration because it can result in a failed-tool flag in Claude's routing logic.
- 2
Add JSON-LD Offer schema with explicit priceValidUntil, priceCurrency, and price fields to every pricing or service page, then re-submit those URLs to Perplexity's index via their Publisher Portal. Faro's /tools/ai-schema generates ready-to-paste structured data for service businesses in under two minutes — prioritize pages where you want to appear in 'best [service] in [city]' answer blocks.
- 3
Rewrite your robots.txt to give GPTBot and PerplexityBot explicit path-level access to your llms.txt file, pricing page, and primary service pages, while disallowing tag archives, author pages, and paginated indexes. Then generate or refresh your llms.txt using Faro's /tools/llms-txt so that when these crawlers land on it, they receive a clean, structured business summary that feeds directly into recommendation context.