Anthropic Embeds MCP Into Claude.ai, Reshaping Business Discovery
Anthropic's move to make MCP a native, always-on layer in Claude.ai shifts AI discoverability from passive indexing to active agent querying — businesses without exposed MCP endpoints are now structurally invisible to a growing share of Claude interactions. Meanwhile, llms.txt tooling is maturing fast and crawler policy divergence is creating new compliance risks.
This Week's Signals
Anthropic Makes MCP Native to Claude.ai, Opening Direct Business Query Layer
Why it matters for your score
With MCP baked into Claude.ai rather than relegated to API integrations, Claude can now query business endpoints directly during a conversation — bypassing traditional web indexing entirely. If your business doesn't expose an MCP-compatible endpoint, Claude has no structured way to retrieve your real-time pricing, availability, or service details. This is the most significant structural shift in AI discoverability since GPTBot launched: it rewards businesses with agent-facing APIs and penalizes those relying solely on crawled web pages.
llms.txt Spec Gains Formal Validator Tooling as Enterprise Adoption Accelerates
Why it matters for your score
The llms.txt specification now has community-maintained validator tooling that checks structural correctness, link resolution, and AI-agent readability — raising the bar from 'file exists' to 'file actually works.' Early adopter data shows AI assistants that support llms.txt are more likely to cite businesses that include specific service descriptions and pricing signals in their files. A malformed or vague llms.txt is increasingly worse than none, because it signals low-quality structured data to crawlers.
Perplexity Confirms Tiered Citation Model: Structured Sources Rank Above Crawled Text
Why it matters for your score
Perplexity has confirmed internally — and acknowledged in developer documentation — that its citation ranking now applies a structured-data preference layer, meaning businesses with valid JSON-LD schema, consistent NAP (name/address/phone) signals, and machine-readable pricing are surfaced ahead of competitors relying on unstructured prose. For local and service businesses especially, this makes schema hygiene a direct revenue variable, not a technical nicety. Businesses without LocalBusiness or Service schema validated against current spec are likely already losing recommendation share.
What to do this week
- 1
Audit your MCP exposure gap this week: if you have no agent-facing endpoint, you are not queryable by Claude.ai's native MCP layer during live user sessions. Use Faro's /tools/ai-readiness-scan to identify whether your business is MCP-addressable and get a prioritized remediation checklist before this gap widens further.
- 2
Regenerate or validate your llms.txt using Faro's /tools/llms-txt — the new validator standard means older files may now fail structural checks. Specifically ensure your service descriptions include pricing signals and category language that matches how AI assistants phrase recommendation queries (e.g., 'best [service] in [city] under $X').
- 3
Run your JSON-LD schema through Faro's /tools/ai-schema against the Perplexity-confirmed structured-source preference criteria: validate that LocalBusiness or Service type, geo coordinates, priceRange, and aggregateRating are all present and correctly nested — then cross-check your robots.txt via /tools/robots-txt to confirm PerplexityBot and ClaudeBot are explicitly permitted, not just defaulting to an unaddressed wildcard rule.