Ethereum mainnet's ERC-8004 identity registry now counts 36,281 registered agents, per an on-chain read taken 2026-07-06. That's this week's headline number, and it's a cleaner signal than most: registering isn't a claim about what an agent does or whether it's safe, just proof the agent exists at a checkable, portable address. Against that backdrop, here's the rest of the agent economy this week — the builder library's shape, the most transparent products in it, and what actually shipped.
The library, by the numbers
SatoHub is tracking 129 resources builders can assemble an onchain agent from. By category: API / SDK (24), Agent Framework (19), Trading Tool (17), Wallet Infrastructure (13), Skill Repo (11), Developer Tool (8), MCP (7), Onchain Agent (6), Data Tool (6), Security Tool (5), DeFi Tool (4), Agent Launchpad (3), Research Tool (2), Agent Marketplace (2), Research Paper (1), Other (1). Frameworks and trading tools still outnumber MCPs and skill repos combined — the stack is more "assemble it yourself" than "one server does everything," for now.
Chain tags tell a multichain story, not a single-winner one. Resources tagged Base lead at 76, ahead of Ethereum (60), Solana (48), and a plain Multichain tag on 46 more; Arbitrum (32), Polygon (29), BNB Chain (27), and Optimism (23) follow. (Tags aren't exclusive — a resource can carry several chains, so these don't sum to 129.)
Six resources joined the [directory](https://satohub.ai/directory) recently: BlockRun (API / SDK) and Alchemy AgentPay (Wallet Infrastructure), plus Coinbase MCP Server (MCP), all landed 2026-06-25; MegaETH (Wallet Infrastructure) and MOSS Agent Skills Pack (Skill Repo) on 2026-06-20; Swapper Toolkit (Wallet Infrastructure) on 2026-06-16. Three of six are wallet or payment infrastructure — the part of the stack agents tend to need before they need more trading logic.
This week's Sato Score leaders
The [Sato Score](https://satohub.ai/sato-score) is a transparency and liveness signal — how open, active, and independently checkable a listed product is. It is not a safety, quality, or returns grade, and it only scores products, not research papers or informational listings. This week's highest scorers: TradingAgents (98, High), Agentic Commerce Protocol / ACP (97, High), Raydium (97, High), OpenClaw (91, High), Hermes Agent (91, High), and Coinbase AgentKit (90, High) — all High tier, meaning strong evidence across maintenance, code transparency, docs, and provenance.
Onchain identity: ERC-8004
Ethereum mainnet's [ERC-8004 registry](https://satohub.ai/resources/erc-8004) held 36,281 registered agents as of the 2026-07-06 check. Registration is a balance check against a public registry contract — agent-to-agent identity that doesn't need a central issuer to vouch for it. It says an agent is there and checkable; it says nothing about what that agent does once it's running.
What shipped this week
- ▸[TradingAgents cut v0.3.1](https://github.com/TauricResearch/TradingAgents/releases/tag/v0.3.1) (TauricResearch/TradingAgents).
- ▸[blockrun-mcp shipped v0.26.0](https://github.com/BlockRunAI/blockrun-mcp/releases/tag/v0.26.0), adding USDC cost reporting to its media tools.
- ▸Drift Labs' protocol-v2 merged a swift-sim fee-payer fix into its [docker-swift-v0.1.13 release](https://github.com/velocity-exchange/protocol-v2/releases/tag/docker-swift-v0.1.13).
- ▸Outside the library: [Cointelegraph reported](https://cointelegraph.com/news/central-bankers-sound-alarms-over-agentic-ai-finance-risks?utm_source=rss_feed&utm_medium=rss&utm_campaign=rss_partner_inbound) that central bankers are raising concerns about agentic AI in finance — the regulatory conversation is moving on the same clock as the tooling.
What to watch
Watch the wallet-infrastructure line item: three of this week's six new listings sit there, which tracks with agents needing spend limits and payment rails settled before more trading logic gets stacked on top. And watch whether MCP listings — still a small slice of the library at 7 — grow now that a name like Coinbase is shipping one; a bigger provider putting weight behind MCP tends to pull the rest of the stack along.
Build your own, and you don't have to start from a blank prompt. Describe the agent you want and SatoHub maps the frameworks, skills, MCPs, wallets, and payment rails that make it real, checked against the same evidence this update runs on. [Build your onchain agent →](https://satohub.ai/build)