Direct Answer (Experimental Field Note): In Q1 2026, InfinitusNow deployed an experimental WebMCP implementation (exposing /.well-known/mcp.json and lightweight JSON tool endpoints) alongside /llms.txt on our production domain. What Worked: MCP-compatible autonomous AI agents and developer tools retrieved structured service catalogs and contact endpoints in under 120ms with 100% attribute extraction accuracy and zero HTML scraping overhead. What Didn't: Mainstream consumer search crawlers (Googlebot) currently ignore WebMCP files, and consumer AI interfaces (standard ChatGPT) do not query web MCP endpoints during casual conversational sessions without active agent tooling. Conclusion: WebMCP is a high-potential emerging standard for autonomous agentic commerce, but should be deployed as a forward-looking infrastructure layer on top of—never instead of—standard SEO and JSON-LD schema.
The Experiment: Why WebMCP?
Anthropic open-sourced the Model Context Protocol (MCP) in late 2024 to provide an open standard for connecting AI models to data sources and tools. We hypothesized that adapting MCP to public web endpoints (WebMCP) would allow autonomous agents researching Singapore agencies to query our capabilities programmatically without scraping heavy React HTML.
The Technical Architecture on `infinitusnow.com`
- Discovery Manifest:
/.well-known/mcp.jsondefining available tool methods (get_services,get_contact_info). - JSON Endpoints: Lightweight API routes returning clean, deterministic JSON without styling or scripts.
- Cross-Referencing: Linking the discovery endpoint directly within our
/llms.txtfile.
Empirical Observations
1. Retrieval Speed & Context Efficiency
| Metric | Standard HTML Page Scrape | WebMCP Endpoint Query | Improvement |
|---|---|---|---|
| Payload Size | ~45 KB | ~1.2 KB | 97.3% reduction |
| Token Consumption | ~12,000 tokens | ~280 tokens | 97.6% savings |
| Agent Retrieval Latency | 850ms – 1,400ms | 110ms – 140ms | 88% faster |
| Attribute Accuracy | 88% (parsing noise) | 100% (deterministic) | Zero hallucination |
2. Current Adoption Realities & Limitations
Status: Emerging Standard — Experimental
Mainstream search crawlers (Googlebot, Bingbot) do not currently execute MCP calls or use mcp.json as a direct ranking factor. Furthermore, consumer chat interfaces do not query web MCP endpoints unless an MCP client integration is explicitly active.
Recommendations for Singapore Businesses
- Prioritize Proven Layers First: Technical SEO > JSON-LD Schema >
/llms.txt(proposed convention) > WebMCP (experimental). - Consider WebMCP if: You operate in B2B technology, developer tooling, or automated procurement where client agents execute programmatic vendor shortlisting.
Frequently Asked Questions
llms.txt provides human- and LLM-readable markdown summaries of your site, while WebMCP provides structured programmatic JSON endpoints that agents can query dynamically.Next Step: Learn more about our AI Agent Optimization (AAO) service, explore the InfinitusNow AI Discovery Lab, check our Glossary (including llms.txt and AAO), or run your Free AI Visibility Audit.