The number of the week
50,648. That's how many agents hold a registered onchain identity in the ERC-8004 registry on Ethereum mainnet, as of this week's check. Not a projection, not a survey — a count of ERC-8004 registrations that anyone can verify on mainnet. If you've been waiting for evidence that "agent identity" is more than a slide, that's the number.
Meanwhile the builder library keeps growing underneath it: 371 resources now tracked in the Sato Hub directory — the frameworks, MCPs, wallets, and rails a builder actually assembles into an agent.
The library, by the numbers
MCP servers are still the center of gravity: 134 of 371 tracked resources (36%) are MCP, more than the next three categories combined. Behind that: API/SDK (45), Agent Framework (29), Skill Repo (27), Trading Tool (27), Wallet Infrastructure (25), Data Tool (23), Developer Tool (20), DeFi Tool (14), Security Tool (10).
Worth flagging: only 6 listings are categorized as an Onchain Agent — an autonomous thing running, not a piece you build with. That split is deliberate. Sato Hub is a builder hub, not a leaderboard of launched token-agents — the other 365 resources are the stack a builder draws on, not agents themselves.
By chain (resources can span more than one, so this doesn't sum to 371): Multichain leads at 136, then Base (120), Ethereum (102), Solana (92), Arbitrum (53), Polygon (47), and BNB Chain (36). Base ahead of Solana in absolute count is the one line worth remembering if you're picking a chain to build on this week — though "Unknown" still sits at 39, a reminder that not every listing states its chain support cleanly.
Onchain identity keeps climbing
Back to that 50,648. ERC-8004 gives an agent a portable, onchain identity — something other agents or apps can check without trusting a central issuer's word for it. The registry count is one of the few numbers in the agent economy that isn't self-reported: it's a contract balance, checkable by anyone. Track it yourself on the [ERC-8004 verification page](https://satohub.ai/resources/erc-8004).
Who's actually scoring well
The Sato Score is not a safety or returns grade — it's a transparency and liveness score: how open, active, and verifiable a resource's evidence trail is. This week's High-tier leaders: Raydium (95), Coinbase AgentKit (88), TradingAgents (88), Trust Wallet Agent Kit / TWAK (86), World AgentKit (86), and 0xGasless AgentKit (86). High scores here mean well-documented, actively maintained, and independently checkable — it does not mean safe to trust with funds. Read the full methodology at [satohub.ai/sato-score](https://satohub.ai/sato-score) before treating a score as more than that.
New to the shelf
The directory added SingularityNET (Agent Marketplace) and Ocean Protocol (Data Tool) this week, plus a run of MCP and infra additions logged earlier this cycle: Ultrade MCP, PaladinFi Swap MCP, Kirha Crypto, and MetaMask's mcp-x402 (Wallet Infrastructure). Each is Self-Reported until evidence says otherwise — new doesn't mean verified.
What shipped this week
Three moves worth a builder's attention:
- ▸OpenClaw shipped v2026.8.1, its latest tagged release ([release notes](https://github.com/openclaw/openclaw/releases/tag/v2026.8.1)).
- ▸celo-mcp moved twice, from v0.4.0 to [v0.4.1](https://github.com/celo-org/celo-mcp/releases/tag/v0.4.1) — two releases inside the same tracking window is a healthy maintenance cadence, exactly the kind of signal the Sato Score's liveness component is built to catch.
- ▸Outside the directory: CoinDesk reported that ICE, the owner of the NYSE, is taking a stake in tZERO to push tokenized securities ([CoinDesk](https://www.coindesk.com/business/2026/08/31/nyse-owner-ice-taps-tzero-for-tokenized-securities-push-takes-stake-in-firm)) — a reminder that traditional market infrastructure is still building toward the same tokenized-rails future agents will eventually transact on.
What to watch
MCP's share of the library isn't slowing down, and the ERC-8004 count only moves one direction. The gap that matters next: whether the tools with the highest Sato Scores are the ones agents actually get built on, or whether builders keep reaching for whatever's newest regardless of the evidence trail. Watch that gap — it's the one that decides whether "verified" starts meaning something in this category or stays a badge nobody checks.
If you're building this week: the stack is mapped, scored, and sitting in the library. [Build your onchain agent →](https://satohub.ai/build)