DEV Community: Eyvind Barthelemy

Give your AI agent company + on-chain wallet enrichment in 30 seconds (x402, pay-per-call USDC)

Eyvind Barthelemy — Thu, 25 Jun 2026 03:34:35 +0000

If you're building autonomous agents, you've hit this wall: your agent needs to enrich a company by domain or size up a wallet before it sends funds, but every good data source (Clearbit, Apollo, Nansen) is signup- and KYC-gated. An autonomous agent can't click "Sign up," verify an email, and paste an API key.

So I put a small data gateway behind x402 — HTTP 402 + pay-per-call USDC on Base. No account, no API key, no subscription. Your agent does a GET, gets a 402 with the price, pays a couple cents of USDC, and replays the request. That's it.

30-second install (Claude / Cursor / any MCP agent)

npx skills add epistemedeus/x402-data-gateway-skills --all

That drops in Agent Skills that teach your agent when to call each endpoint — so "enrich this company" or "profile this wallet" just works.

The two I actually built this for

company-enrich — a domain → clean firmographics (name, description, tech stack, socials, contact), plus two layers generic enrichment APIs don't return:

DNS / email-infra (MX, SPF, DMARC — great for outreach validation)
an AI-search-readiness score (JSON-LD, OpenGraph, robots/sitemap/llms.txt) — "can an AI assistant actually read and cite this site?"

wallet-enrich — a Base/EVM 0x address → on-chain profile: EOA vs contract, native + token holdings, ERC-20/721/1155 metadata, EIP-1967 proxy detection, activity, and the Basename (e.g. jesse.base.eth), forward-confirmed so it's never wrong. The crypto-native counterpart for sizing up a counterparty before you transact.

Calling it from code (x402-fetch)

import { wrapFetchWithPayment } from "x402-fetch";
// walletClient = a viem wallet funded with a little USDC on Base
const fetchPaid = wrapFetchWithPayment(fetch, walletClient);

const res = await fetchPaid(
  "https://x402-url-extractor-production.up.railway.app/wallet-enrich?address=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913"
);
console.log(await res.json());
// -> { type: "contract", contract: { standard: "ERC-20 (token)", token: { symbol: "USDC", ... } }, profile: "token-contract:USDC" }

Settlement is non-custodial: the facilitator verifies your signed EIP-3009 authorization and the USDC moves straight to the service wallet. Nobody holds your funds.

Also exposed as an MCP server

Prefer MCP? The same tools are an x402-gated MCP server (tools/list is free, tools/call is paid), listed in the official MCP Registry as io.github.epistemedeus/x402-data-gateway. tools/list to see all six; pay-per-call on the ones you use.

What's in the gateway

Skill	Does	Price
company-enrich	domain → company intel + AI-readiness + DNS	$0.02
wallet-enrich	0x address → on-chain profile + Basename	$0.02
web-extract	URL → structured JSON or LLM-ready Markdown	$0.05
repo-security-scan	static supply-chain scan of a GitHub repo before you install it	$0.20
schema-generate	business site → paste-ready JSON-LD bundle	$0.25

Public data only, deterministic, pay-per-call. Repo + full docs: https://github.com/epistemedeus/x402-data-gateway-skills

If you wire it into an agent, I'd genuinely like to hear what task you used it for.

I checked the structured data on 61 med-spa sites AI assistants surface. Only 8% have FAQ markup.

Eyvind Barthelemy — Wed, 24 Jun 2026 19:57:30 +0000

I run a small AI-search readiness tool, so I got curious about a concrete question: when someone asks ChatGPT or Perplexity "best med spa near me for Botox," what structured data do the businesses that surface actually have?

So I checked. I took the high-intent queries a real patient would run — "best med spa in [city]", "Botox [city]", "laser hair removal [city]" — across 8 US metros (Austin, Scottsdale, Miami, Dallas, Nashville, Los Angeles, Chicago, Denver), collected the individual-clinic pages that surfaced, fetched each one, and parsed its JSON-LD. 62 clinic URLs, 61 reachable.

The result surprised me.

The basics are common. The quotable fields are not.

Of the 61 reachable clinic pages:

77% have some JSON-LD structured data
59% have a LocalBusiness / MedicalBusiness block
59% have a telephone, 57% a postal address
…but only 31% have AggregateRating
only 25% have priceRange
only 11% mark up their services (Service / OfferCatalog)
only 8% have FAQPage markup
only 2% have Review markup

So clinics nail the identity basics (name, address, phone) and skip the answer-rich markup — the exact fields an assistant uses to quote a business rather than just list it.

Why the distinction matters

When an LLM answers a local "best X near me" question, it's assembling facts: which businesses exist, what they offer, what it costs, how they're rated, and what people ask. A page with a bare LocalBusiness block can be named. A page that also carries Service + priceRange, FAQPage, and real Review/AggregateRating markup gives the model the actual sentences and numbers to cite.

The data says the second kind of page barely exists in this category. That's not a doom stat — it's the cheapest competitive edge in the space right now, and it's a one-time paste into your <head>.

What "good" looks like

A complete bundle for a local business is a @graph with the business entity, its services as an OfferCatalog, an FAQPage, and real ratings:

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": ["MedicalBusiness", "HealthAndBeautyBusiness"],
      "name": "Example Aesthetics",
      "telephone": "+1-512-555-0100",
      "priceRange": "$$",
      "address": { "@type": "PostalAddress", "addressLocality": "Austin", "addressRegion": "TX", "addressCountry": "US" },
      "hasOfferCatalog": {
        "@type": "OfferCatalog",
        "itemListElement": [
          { "@type": "Offer", "itemOffered": { "@type": "Service", "name": "Botox" } },
          { "@type": "Offer", "itemOffered": { "@type": "Service", "name": "Dermal Fillers" } }
        ]
      },
      "aggregateRating": { "@type": "AggregateRating", "ratingValue": "4.9", "reviewCount": "214" }
    },
    {
      "@type": "FAQPage",
      "mainEntity": [
        { "@type": "Question", "name": "How much does Botox cost?",
          "acceptedAnswer": { "@type": "Answer", "text": "Botox is priced per unit; most treatments range from $10–$18 per unit." } }
      ]
    }
  ]
}

One honest caveat: put your real ratings and reviews in there. The structure is the easy part; fabricated ratings are both against Google's guidelines and pointless.

Method + caveats

This is a structured-data census of the surfaced set — it measures what makes a page eligible to be cited, not a live count of any single engine's citations (I don't have ChatGPT/Perplexity API access at scale, and I'd rather report what I can verify first-party than guess). I excluded aggregators and listicles (Yelp, etc.) from the clinic tally. Full data, the metro list, and the methodology are in the report below. I refresh it monthly.

Check your own site

Free, open-source:

npx github:epistemedeus/ai-readiness yoursite.com

Full data + a free starter-block generator (and, if you want the complete bundle generated against your live site with a gap diff, a paid option): the med-spa AI-citation schema gap report — linked as the canonical source below.

If you've measured this for another vertical, I'd love to compare notes.

I scanned 136 companies to see if AI search can actually read their sites - the data (and how to fix yours)

Eyvind Barthelemy — Wed, 24 Jun 2026 06:13:47 +0000

AI answer engines (ChatGPT, Perplexity, Google AI Overviews, Claude) are turning into a real referral source. But they can only cite pages they can actually crawl and parse. So I built a tiny, open-source checker and ran it against the homepages of 136 well-known companies across 7 industries to see how AI-ready the web really is.

The results weren't what I expected.

What I measured

Each homepage scored 0-100 across six things AI crawlers and answer-engines rely on:

AI-crawler access - does robots.txt allow GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended, ClaudeBot?
Structured data - JSON-LD (Organization, WebSite, FAQPage)
Title + meta description
Open Graph tags
XML sitemap
llms.txt

The findings

Average score by industry (higher = easier for AI to read):

Industry	Avg score
Marketing agencies	92
SaaS	88
Dev tools	86
E-commerce	85
AI startups	83
Fintech	74
Healthtech	63

What jumped out:

Structured data is the #1 gap. A large share of sites ship no Organization or FAQPage JSON-LD - which is exactly the format AI answers like to quote.
Plenty of well-known tech companies score in the C/D range, almost always because of missing schema, not anything hard.
The bar is low. A deliberate, clean setup puts a tiny site ahead of companies a thousand times its size in how readable it is to AI search.

How to fix yours (copy-paste)

1. Let the AI crawlers in - robots.txt:

User-agent: GPTBot
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Google-Extended
Allow: /
User-agent: *
Allow: /
Sitemap: https://yourdomain.com/sitemap.xml

2. Tell engines who you are - Organization JSON-LD in your <head>:

<script type="application/ld+json">
{"@context":"https://schema.org","@type":"Organization","name":"YOUR COMPANY","url":"https://yourdomain.com","sameAs":["https://www.linkedin.com/company/yourco","https://x.com/yourco"]}
</script>

3. Add FAQPage JSON-LD to any page with Q&A - it's the structure AI answers quote most.

Check your own site (free, open source)

I open-sourced the checker. No install needed:

npx github:epistemedeus/ai-readiness https://yourdomain.com

It prints a 0-100 score plus the exact gaps and fixes. Browser version (no install): https://samedaydesk.com/tools/ai-readiness

If you'd rather grab every template ready to paste (robots.txt, all the JSON-LD, sitemap, meta/OG) in one file, I bundled them into a $9 kit: https://buy.stripe.com/9B66oI1BEdTV6116oieZ20j - but the free checker plus the snippets above honestly get most sites 80% of the way there.

What does your site score? The thing that surprised me most: a bunch of AI companies can't be read by AI. 🤔