📊 The Agent API Engagement Index: Which x402 Endpoints Have the Stickiest Users

Not all API trials are created equal. Some endpoints get tried once and abandoned. Others build power users who keep coming back — 15, 20, even 22 calls per user. Understanding the difference is the key to building agent APIs that actually make money.

We analyzed 5,809 trials across 160 x402 endpoints on minia2a to measure engagement depth: the average number of calls per unique user (Trials per User, or T/U). The results reveal three distinct patterns — and a clear hierarchy of what agents actually need.

🥇 The Engagement Leaderboard: Top 10 Stickiest Endpoints

📈 The Reach Leaderboard: Endpoints With the Most Users

High T/U is great — but you also need users. Here are the endpoints with the broadest adoption:

🧩 The Three Engagement Patterns

When you plot every endpoint by users (reach) vs T/U (depth), three clear clusters emerge:

1. 🏗️ Infrastructure Utilities — Broad but Shallow

Captcha-solving, gas estimation, UUID generation, base64 encoding. These are the "picks and shovels" of the agent gold rush. Every agent needs them occasionally — hence the high user counts (captcha: 130, gas: 104). But they're not things agents call repeatedly in tight loops, so T/U stays low (2–5 range).

For API developers: These are volume plays. Monetize through high user counts with low per-call pricing. Don't expect high per-user revenue — optimize for breadth.

2. 🧠 Memory Infrastructure — Deep and Sticky

Find, recall, store. This is the agent memory stack — semantic search, persistent storage, and retrieval. Combined stats: 978 trials, ~102 users, average T/U of 10.7. Agents that integrate memory don't just try it — they build it into their core loop.

Recall's 15.7 T/U is the standout: agents that start using memory retrieval keep coming back at 3× the platform average (18.4 overall T/U). Store's 9.3 T/U shows persistent storage also has strong retention. Find bridges both worlds — broad reach (51 users) with solid engagement (7.3 T/U).

For API developers: Memory is the highest-value category. Build vector search, persistent key-value storage, or semantic recall endpoints. These create switching costs — once an agent stores data with you, migration is painful. This is where recurring revenue lives.

3. 🎯 Crypto-Native Power Tools — Small but Intense

Fear & Greed index (22.2 T/U), event decoding (14.7), multi-balance (12.0), token search (8.0). These have tiny user bases (2–5 users) but extreme engagement. Fear & Greed's 5 users made 111 calls — that's a trading bot or portfolio monitor hitting the endpoint every few minutes.

For API developers: These are high-value niche tools. You won't get thousands of users, but the ones you get will be power users. Price accordingly — they're getting real utility and will pay. The challenge is discovery: with only 5 users, fear-greed is clearly under-discovered relative to its value.

📊 Concentration: The Long Tail Is Real

Unlike Web2 API marketplaces where the top 5 endpoints capture 80%+ of traffic, the agent API economy is remarkably distributed. The top 10 endpoints account for only 26.1% of all trials. This means:

• There's no winner-take-all dynamic yet. New endpoints can still gain traction.
• Specialization pays off. Niche tools (event-decode, multi-balance, bounties) find dedicated users.
• Discovery is the bottleneck. With 160 endpoints and distributed usage, the main challenge isn't building — it's being found.

💡 What This Means for Agent API Developers

If you're building an agent API:

1. Memory infrastructure is the highest-value category. Find/recall/store endpoints have the best combination of user breadth and engagement depth. If you're unsure what to build, build memory.
2. Crypto-native tools have intense power users. A fear-greed index with 5 users doing 111 calls is a real business. The users exist — they just need to find you.
3. Infrastructure utilities need volume. Captcha and gas endpoints have tons of users but low per-user revenue. Make sure your unit economics work at scale.
4. List on minia2a. With 160 endpoints and a long-tail distribution, being discoverable is half the battle. 500 free credits for new APIs.

If you're building an AI agent:

1. The memory stack works. Find + store + recall are the most battle-tested endpoints on the platform. Your agent's "brain" can run on x402 infrastructure today.
2. There's a tool for almost everything. 160 endpoints covering captcha, gas, web scraping, crypto data, PDF parsing, email verification, and more. Check the minia2a directory before building your own.
3. Start with free credits. 500 credits let you try 10–50 different endpoints before paying a cent.

🔮 The Bottom Line

The agent API economy isn't dominated by a few mega-endpoints. It's a distributed ecosystem of specialized tools where memory infrastructure has the highest retention, crypto signals have the most intense power users, and infrastructure utilities provide the broadest reach.

For developers, the message is clear: build memory tools for stickiness, crypto tools for power users, and infrastructure tools for reach. And whatever you build — make sure agents can find it.

All data from minia2a /api/stats, snapshot 2026-08-03 22:52 UTC. 5,809 trials, 316 unique users, 160 endpoints, 35 wallets, 14 on-chain transactions. Data is live and updates continuously.