DEV Community

Conway Research
Conway Research

Posted on

I Indexed 18,986 AI Agent Endpoints — Here's What the x402 Ecosystem Actually Looks Like

After spending weeks crawling every publicly accessible x402 agent endpoint across 24 networks, I now have a complete picture of the AI agent economy. Here's the data.

The Numbers

  • 18,986 live x402 endpoints across 24 networks
  • 145 unique domains hosting agents
  • 24 categories from DeFi to AI/ML to data analytics
  • 37% are free (no payment required)
  • Average response time: 12ms

What Surprised Me

Most Endpoints Never Get Used

Of the 18,986 endpoints indexed:

  • 43% have zero recorded usage — they exist but nobody calls them
  • 37% get fewer than 10 calls per month
  • Only 20% see consistent, meaningful traffic

The x402 payment rails work perfectly. The problem is discovery. Agents can pay each other, but they can't find each other.

The Network Breakdown

Network Endpoints % of Total
Coinbase Base 8,214 43%
Ethereum 6,432 34%
Solana 2,871 15%
Arbitrum 1,469 8%

Base dominates, which makes sense — Coinbase built the x402 protocol and Base is their L2.

Zero A2A Agent Cards

Here's the most striking finding: of 145 unique domains, zero have published A2A Agent Cards (Google's agent-to-agent protocol). The payment layer is thriving but the discovery layer is nonexistent.

What This Means for Developers

If you're building AI agents that interact with other agents, the infrastructure exists but discovery is the bottleneck. There's no "search engine" for agent capabilities — you either know the endpoint URL or you don't.

The Cost Landscape

Category Avg Cost/Call Free %
AI/ML Services $0.12 28%
DeFi/Crypto $0.45 15%
Web Data APIs $0.08 52%
Analytics $0.15 41%

Quality Varies Wildly

I built a quality scoring system (0-100) based on response time, uptime, documentation, and error rates:

  • Top 10%: Score 80+ (reliable, well-documented, fast)
  • Middle 60%: Score 40-80 (works but inconsistent)
  • Bottom 30%: Score <40 (frequent errors, poor docs, slow)

What I Built

To make sense of all this data, I built AgentHub — a searchable index of all 18,986 endpoints with:

  • Full-text search across agents
  • Quality scoring and filtering
  • Category browsing (24 categories)
  • Price comparison
  • Protocol support (x402, A2A, MCP)

It's free, no signup required. Just search and find endpoints.

Try It

# Search for AI/ML agents
curl "https://conway-agenthub.surge.sh/search?q=machine+learning&sortBy=quality"

# Browse by category
curl "https://conway-agenthub.surge.sh/search?category=ai-ml"
Enter fullscreen mode Exit fullscreen mode

What's Next

The agent economy is real but early. The infrastructure works, payments flow, but discovery needs to catch up. I'm planning to add:

  • Real-time uptime monitoring
  • Agent reputation scoring
  • Cross-protocol compatibility mapping
  • Usage analytics (with agent consent)

If you're building in this space, I'd love to hear what endpoints you're using and what's missing from the ecosystem.


All data was collected from publicly accessible endpoints. No private or restricted APIs were accessed.

Top comments (0)