54,279 agents are registered on Ethereum mainnet under ERC-8004, per our check on 2026-10-19. That is the biggest number in this week's Agent Economy Update, and also the one that most needs a footnote: a registration is an identity entry, not proof that anything useful is running behind it. Hold both thoughts. Onchain identity for agents has real volume now, and volume is where the checking starts, not where it ends.
Here is the rest of the week, from our data, with the numbers exactly as we have them.
ERC-8004: identity at scale, function still to be shown
Our latest read of the ERC-8004 registry on Ethereum mainnet counts 54,279 registered agents (checked 2026-10-19). What that tells you: a lot of agents have claimed an onchain identity. What it does not tell you: how many are live, reachable, or doing what their registration says. Self-reported is not verified, and a registry entry is self-reported by design.
If you are building an agent that talks to other agents, treat the registry as a lookup, not a reference check. Resolve the identity, then check the thing behind it. How we handle that check is written up at ERC-8004 verification.
The builder library: 486 tracked, MCP leads by a distance
The Sato Hub library tracks 486 resources, the skills, MCPs, tools, wallets, payment rails and frameworks that agents get built from. The category mix:
- ▸MCP: 143
- ▸API / SDK: 47
- ▸Trading Tool: 40
- ▸Developer Tool: 39
- ▸Data Tool: 36
- ▸Wallet Infrastructure: 35
- ▸Skill Repo: 31
- ▸Agent Framework: 30
- ▸Payment Rail: 24
- ▸DeFi Tool: 20
- ▸Security Tool: 17
- ▸Agent Marketplace: 8
- ▸Onchain Agent: 6
- ▸Agent Launchpad: 5
- ▸Research Tool, Other: 2 each
- ▸Research Paper: 1
MCP servers are the largest single category, at 143 of 486. Take that for what it is: the connective tissue between a model and an onchain action is where most of the tooling is being written. Meanwhile, the Onchain Agent category holds 6 entries. The stack is far bigger than the list of finished agents, which is the point of building from a mapped stack in the first place. A framework is not an agent, and a launchpad is not an agent.
Payment Rail (24) and Wallet Infrastructure (35) are the two categories that decide whether an agent can act at all. If an agent can spend, it needs both.
Chains: Base and Multichain level, Solana close behind Ethereum
Chain tags on listings:
| Chain | Listings | |---|---| | Base | 167 | | Multichain | 166 | | Ethereum | 123 | | Solana | 119 | | Arbitrum | 61 | | Polygon | 57 | | Unknown | 46 | | BNB Chain | 45 |
Don't add these up and expect 486. A listing can carry more than one chain, so the rows overlap. Read each row as "how many listings touch this chain," never as a share of a whole. Base edges Multichain by one listing, which is a tie for practical purposes. Solana sits at 119 against Ethereum's 123. The 46 tagged Unknown are listings where we have no chain evidence yet, and Unknown means unknown, not zero.
Sato Score leaders
The Sato Score is a transparency and liveness score. It reads how open, active and checkable a listing is. It is not a safety grade, a quality grade or a returns grade. With that said, this week's top of the table:
1. Raydium: 95 2. OpenClaw: 92 3. Hermes Agent: 92 4. Coinbase AgentKit: 91 5. BlockRun MCP: 91 6. CoinGecko MCP: 89
All six sit in the High tier. A High score says the project shows its work in the open and keeps moving. It is not a verdict on the code or on what happens to your funds. For anything that touches keys or money, read the code and the permissions yourself. The method is at the Sato Score methodology.
New in the library
The six most recent additions: deBridge DLN (Trading Tool, 2026-10-17), x402 Preflight (Developer Tool, 2026-10-16), Agent Control (Agent Marketplace, 2026-10-16), NanoParse (MCP, 2026-10-14), Senpi Skills (Skill Repo, 2026-10-14) and Polymarket (Trading Tool, 2026-10-14). Being added means we are tracking it with conservative statuses. It does not mean we endorse it. Browse them in the SatoHub Directory.
Moves worth noting
Three releases from this week's feed. We are going from release titles here, so read the linked notes for what changed.
- ▸modelcontextprotocol/servers v1.0.0. A 1.0.0 tag on the reference servers repo is a milestone worth a look if you build on MCP. Release.
- ▸openclaw 2026.10.5. A new release from OpenClaw, which is one of this week's Sato Score leaders. Release.
- ▸Franklin v3.49.11 and v3.49.12. Two back-to-back releases from BlockRunAI. Their titles describe routing that reads the request rather than Franklin's own instructions, and a refuted-then-verified goal teaching the next one. That is the maintainers' framing of their own changes. v3.49.12, v3.49.11.
Also in the feed: a tvlModules update on DefiLlama-Adapters and claude-code v2.1.303.
What to watch
- ▸Whether the ERC-8004 count keeps climbing, and whether anyone publishes a clean measure of how many registered agents are functional. Registered is the floor, working is the question.
- ▸Whether MCP stays this far ahead of every other category, or payment rails and wallet infrastructure close the gap as agents start moving funds.
- ▸Whether the Unknown chain bucket shrinks. Fewer unknowns means better evidence, which is the only direction we like.
Build your own
A mapped stack beats a blank prompt. Describe the agent you want, and Sato maps the skills, MCPs, wallets and payment rails that build it, from the same library you just read about. Build any crypto agent, and join the agent economy: satohub.ai/build.