Claude (Anthropic) prefers clear main content, transparent sources and tooling-friendly markup via the Model Context Protocol (MCP). SGE-SCORE audits 12 Claude-relevant checkpoints (HTML cleanliness, source transparency, MCP availability); the Master-Fix produces the necessary patches including MCP-compliant endpoints.
How Claude processes content
Claude (Sonnet, Opus) has used a web-browse capability since late 2024 and added tool-use over the open Model Context Protocol (MCP) in 2025. Unlike ChatGPT and Gemini, Anthropic explicitly tunes the model for safe sourcing — visible sources and traceable arguments beat raw answer speed.
In practice this means: a page with a clear heading structure, a visible author and external linking to primary sources is preferred inside Claude. Domains with an MCP endpoint (/.well-known/mcp.json) gain extra weight because Claude recognises them as "tool-capable" sources — see modelcontextprotocol.io.
"Claude weighs source transparency higher than any other engine — and additionally rewards MCP availability."
→ Next step: Let our free demo analysis also check your MCP setup.
The 7 most important signals for Claude visibility
- Clean main content with no aggressive overlays, modals or tracker scripts before the content.
- Source transparency: external links to primary sources with
rel="noopener". - Visible author with Person schema and external profile link.
- MCP endpoint:
/.well-known/mcp.jsonwith described tools / resources. - OpenAPI or plugin manifest — Claude reads both for tool-use.
- Clear heading structure with semantic
section idanchors for quotation. - Update date visible & in schema.
→ Next step: More background in the SGE-GEO knowledge hub.
What our audit checks specifically for Claude
Out of 57 checkpoints (max. 114 points), these 12 are particularly relevant for Claude:
| # | Checkpoint | Sector | Max. pts | Patent / source |
|---|---|---|---|---|
| 11 | Initial HTML contains main content | II — Parse | 2 | US 10,275,452 B2 |
| 15 | Clean H2/H3 structure | II — Parse | 2 | HTML5 sectioning |
| 16 | Section IDs as anchors | II — Parse | 2 | HTML5 best practice |
| 22 | Complete Article schema | III — Schema | 2 | schema.org/Article |
| 25 | Person schema with sameAs | III — Schema | 2 | Knowledge Graph API |
| 31 | Author bio with Person schema | IV — Trust | 2 | E-E-A-T guidelines |
| 33 | Date & last-updated visible | IV — Trust | 2 | QRG, section 5.1 |
| 43 | External high-authority links | IV — Trust | 2 | QRG, section 5.4 |
| 44 | rel="noopener" on external links | IV — Trust | 2 | HTML5 spec |
| 54 | agent.json + ai-plugin.json | V — Agentic | 2 | Agentic Commerce spec |
| 55 | OpenAPI manifest | V — Agentic | 2 | OpenAPI 3.1 |
| 56 | MCP endpoint reachable | V — Agentic | 2 | modelcontextprotocol.io |
Related post: Agentic Commerce & AI Shopping — the Next Stage. Methodology in the Docs section.
Before / after: what the Master-Fix produces concretely
Before (typical MCP state)
// File /.well-known/mcp.json is missing entirelyAfter (Master-Fix output)
// /.well-known/mcp.json
{
"$schema": "https://modelcontextprotocol.io/schema/0.1",
"name": "example.com",
"description": "MCP endpoint for Example AG product knowledge",
"version": "1.0.0",
"endpoints": {
"rest": "https://example.com/api/mcp"
},
"resources": [
{ "name": "products", "description": "All product data" },
{ "name": "docs", "description": "Product documentation" }
],
"tools": [
{ "name": "search_docs", "description": "Search product docs" }
]
}Plus: matching Article / FAQPage / BreadcrumbList schema, author bio with Person schema, clean noopener on all external links — as one patch set.
→ Next step: Master-Fix promo: 9 EUR.
Using MCP & tool-use correctly
The Model Context Protocol (MCP) was proposed by Anthropic in 2024 and is now supported by OpenAI & Google. It describes how an AI model can declaratively address a domain's tools and data sources — without each provider having to register its own plugin.
- Required file:
/.well-known/mcp.json - Recommended: OpenAPI spec at
/openapi.yaml - Recommended:
/.well-known/agent.jsonfor agentic discovery
A correct MCP implementation is worth 2 audit points and positions your domain as "tool-ready" — see knowledge hub, sector V (Agentic).
The 5 most common mistakes in Claude optimisation
- Cookie banner covers main content — Claude only reads what's visible.
- External links without
rel="noopener"— lose trust points. - MCP file missing — no tool-use binding.
- Anonymous authors — no source trust.
- No section IDs — Claude cannot quote precisely.
Frequently asked questions (FAQ)
Which Claude models use web browsing?
Currently Sonnet 4.5 and Opus 4.5 (as of April 2026). The browse capability is available in the API via the tool_use feature.
Do I really need an MCP endpoint?
Not strictly required, but strongly recommended. SGE-SCORE awards 2 points for a valid mcp.json — the Master-Fix generates it automatically.
What is the difference to ai-plugin.json?
ai-plugin.json was an OpenAI-specific format. MCP is open and provider-agnostic — we recommend providing both.
What does the Master-Fix cost?
Currently 9 EUR. Output: HTML patches, schema JSON, an mcp.json proposal and an author bio.
How do I measure Claude visibility?
Through patent/heuristic-based proxies, since Anthropic does not provide an official API for citation tracking. SGE-SCORE delivers a consistent comparison baseline.
Does this also work in Claude.ai projects?
Yes. Claude.ai projects use the same web-browse and tool-use mechanisms as the API.
Which external sources count as "high authority"?
Government (gov, eu), universities (edu, ac), standardisation bodies (W3C, IETF), peer-reviewed studies (DOI), and major tech vendor documentation.
Audit your visibility for Claude
Start a free, patent-based analysis. SGE-SCORE checks 57 criteria — including the 12 Claude signals from this post.