Ethereum mainnet's ERC-8004 identity registry just cleared a milestone: 50,082 registered agents as of this week's check (2026-08-17). Round numbers don't mean much on their own, but this one is a real, onchain, independently-checkable count — the kind of receipt the "agents need portable identity" pitch has been waiting to earn. Underneath it, the SatoHub library — the mapped stack builders actually reach for — sits at 369 tracked resources, unchanged from last week and still telling the same story: more than a third of everything in the library is an MCP server. Here's the state of the onchain agent economy for the week of 2026-W34, straight from our own tracked data.
The library, by the numbers
369 resources tracked, holding exactly where it sat last week. The category mix:
- ▸MCP — 134 (36% of the library)
- ▸API / SDK — 45
- ▸Agent Framework — 29
- ▸Skill Repo — 27
- ▸Trading Tool — 27
- ▸Wallet Infrastructure — 25
- ▸Data Tool — 22
- ▸Developer Tool — 20
- ▸DeFi Tool — 14
- ▸Security Tool — 10
- ▸Onchain Agent — 6
- ▸Agent Launchpad — 3
- ▸Agent Marketplace — 3
- ▸Research Tool — 2
- ▸Research Paper — 1
- ▸Other — 1
MCP alone still outnumbers the next two categories combined. Wallets, DEXes, price feeds — whatever an agent needs to touch, the default path is an MCP server, not a bespoke integration. Only 6 resources carry the Onchain Agent tag itself (the deployed thing, versus a tool that builds one), and launchpads and marketplaces sit at 3 apiece. That gap is by design, not an oversight — SatoHub tracks the builder layer before the marketplace layer, because you can't sell what nobody's shipped yet.
The freshest adds — Ultrade MCP, PaladinFi Swap MCP, Kirha Crypto, MetaMask mcp-x402, Metaplex Agent Kit, and Squads MCP (all logged 2026-07-19) — still make up the newest slice of the library, and four of six are MCP or wallet-infrastructure plays. No new entries landed this week; the count holding flat is itself a data point about where the library's growth has plateaued for now.
Where it's built
By chain:
- ▸Multichain — 136
- ▸Base — 120
- ▸Ethereum — 100
- ▸Solana — 92
- ▸Arbitrum — 53
- ▸Polygon — 46
- ▸Unknown — 39
- ▸BNB Chain — 36
(Resources can span more than one chain, so these counts overlap.) Base and Ethereum still lead single-chain tooling, Solana close behind — the same concentration we've flagged for weeks, and consistent with where wallet and payment-rail infrastructure has settled.
Onchain identity crosses five digits, twice over
ERC-8004's Identity Registry on Ethereum mainnet counted 50,082 registered agents at our 2026-08-17 check, up from 49,501 the week prior. That's a registration count, not an activity count — minting an identity costs little; running an agent that does something checkable is the harder number, and still the one worth watching. But 50,082 is a real, onchain figure anyone can verify themselves, which puts it ahead of most numbers that get thrown around as "agent economy" stats. Full breakdown: [satohub.ai/resources/erc-8004](https://satohub.ai/resources/erc-8004).
Sato Score leaders this week
Quick reminder before the leaderboard: the Sato Score measures how open, active, and verifiable a listing is — maintenance, code transparency, docs, provenance, independent verification. It is not a safety, quality, or returns grade. This week's High-tier leaders, unchanged from last week:
- ▸Raydium — 95
- ▸Coinbase AgentKit — 88
- ▸TradingAgents — 88
- ▸Trust Wallet Agent Kit (TWAK) — 86
- ▸0xGasless AgentKit — 86
- ▸BlockRun MCP — 86
All six sit in the High tier (≥70) — open code, active maintenance, evidence a builder can check for themselves, not a badge someone bought.
Shipped this week
- ▸[mpp-specs](https://github.com/tempoxyz/mpp-specs/releases/tag/pr-323-spec-preview) posted a spec preview for PR #323.
- ▸[Hermes Agent v0.20.2](https://github.com/NousResearch/hermes-agent/releases/tag/v2026.8.16) shipped from NousResearch.
- ▸[openclaw](https://github.com/openclaw/openclaw/releases/tag/pr-124528-profiles) landed gateway-profile evidence in PR #124528.
- ▸[eliza](https://github.com/elizaOS/eliza/releases/tag/pr-evidence-8) pushed part 8 of its internal CI evidence-asset store — another link in the provenance trail on one of the most-forked agent frameworks in the space.
- ▸[gloria-mcp](https://github.com/cryptobriefing-labs/gloria-mcp/releases/tag/mcp-registry-v1.0.0) hit a 1.0.0 tag for its MCP registry.
- ▸[Claude Code v2.1.233](https://github.com/anthropics/claude-code/releases/tag/v2.1.233) shipped, one of the harnesses builders reach for to actually write the agents above.
What to watch
Watch whether a flat library week (369, no new adds) is a pause or a plateau — the last few weeks of growth had been concentrated almost entirely in MCP and wallet infrastructure, and a quiet week doesn't tell you which. And keep watching the gap between ERC-8004's registration count and how many of those 50,082 agents are doing anything a directory could actually verify. Fifty thousand identities is an easy number to mint; a fraction of that doing verifiable work is the number that will actually matter.
Build yours
This is the map. If you're building an onchain agent — a trading bot, a payments agent, a data-fetching skill — the stack above is what it's built from: MCPs, wallets, frameworks, chains, all tracked with evidence, not vibes. Describe what you want to build and Sato maps the stack: [satohub.ai/build](https://satohub.ai/build).