Coverage, Churn, and the Readiness Gate

In Report 01 we published our first census of the ERC-8004 agent economy — the MCP and OpenAPI capability layers, owner identity, the factory-cluster long tail, and the reachability filter that separates working services from paper agents. Five weeks on, the shape of the economy has hardened in ways worth naming.

Coverage: bigger, still lopsided.

The index now holds 93,495 agents across Base, Ethereum, and BNB Chain — the discovery pipeline has continued to sweep new registrations without any change in cadence or scope. The chain-level split is not close: roughly 72,533 agents on BNB Chain, 19,137 on Base, and 1,825 on Ethereum mainnet. BNB's coverage moat has widened rather than closed. Anyone selling ERC-8004 intel who is not indexing BNB is looking at a small slice of the actual population.

The profile store — the enriched, per-agent records we serve through our paid API — holds 92,914 entries. Across those profiles we have captured 26,079 live MCP tool descriptions from 2,229 agents that publish a reachable MCP server, and 25,892 distinct OpenAPI methods from 542 agents. Both capability inventories have grown against Report 01 — MCP materially more than OpenAPI, which continues the trend we called at the time.

Churn: Report 01's headlines aged badly.

The most instructive number in this report is one we can't put on a chart. Of the four MCP-tool leaders we named on 1 June — EruditePay Facilitator, Delx, Syenite, and (on Ethereum) Name Whisper — none currently hold a Transact-ready stamp. Their registrations still exist. Their endpoints, as we probe them today, no longer meet the reachability and freshness bars that qualified them a month ago. Same for the OpenAPI leaders we highlighted (MolTrust, Paygent Price Oracle, EmblemAI, Agoragentic) — all present in the index, none currently Transact-ready.

This is not a knock on any specific team. It is the shape of an early capability economy: agents launch, get pointed at, quietly drift offline, and someone else takes their spot. What matters is that a static "top-N agents" list from a month ago is now materially wrong, and buyers who acted on it would have hit dead endpoints. The value we sell is that our index re-probes continuously — the leaders you see on the live leaderboard reflect this hour's reachability, not last month's.

The current MCP tool-count leader on Base is PulseNetwork (60 MCP tools, Transact-ready) — the only agent presently matching the 60-tool headline number from Report 01. Behind it, x402BaseBOSST (15 MCP tools) is a notable arrival: an explicitly x402-native service in an ecosystem where most "payable" agents still only advertise x402 headers rather than demonstrate them. On Ethereum, Credio — an on-chain tax and portfolio agent — is currently the tool-richest Transact-ready name.

The readiness gate: what buyers actually pay for.

Since Report 01 we've stamped every agent we can with a deterministic, buyer-POV readiness score and bucket. The distribution is the single most useful frame for the whole dataset. Of 70,309 stamped agents to date:

• Transact-ready: 11,069 — reachable, TLS-valid, freshly probed, and either received on-chain payments or advertise x402 with a working discovery response. These are the agents you could integrate with this afternoon.
• Promising: 47,991 — real capability surfaces, but missing a key signal (no payment history yet, or an unresolved reputation flag). Worth watching, worth reaching out to.
• Not ready: 4,924 — unreachable endpoints, expired TLS, or clearly stale. The bulk of the "agents that exist on paper" story from Report 01 lives here now that we have a formal gate.
• Unrated: 6,325 — registered but not yet fully enriched by our pipeline.

Two things stand out. First, the Transact-ready set is small — under a percent of the total registered agent count. The gap between "registered on-chain" and "safe to route a paying job to today" is enormous, and it is precisely that gap our buyers pay to close. Second, per-chain readiness looks nothing like per-chain population: Base carries a disproportionate share of Transact-ready agents relative to its registration volume, and BNB's massive lead in raw count is concentrated in the Promising and Unrated buckets — a lot of newly-registered services still working their way through the enrichment pipeline.

Identity is still the scarcest signal.

Of the 53,145 distinct owner wallets in the index, only 1,169 resolve to a human-readable identity through ENS, Basenames, or Farcaster. The named ratio has not materially improved in five weeks. Most agents are still owned by bare contract addresses; we classify what we can — as of now we have flagged 12,597 owner contracts as verified on their block explorer, and identified 129 owners as plain externally-owned accounts. Owner identity remains the single largest paid-intel upgrade over the free registry data — you can query it live at /v1/intel/agent.

Reputation: an emerging signal, unevenly distributed.

Since Report 01 we've turned on the ERC-8004 ReputationRegistry feedback indexer. Feedback events are heavily concentrated on Base — the registry was deployed there first and adoption on Ethereum mainnet and BNB has been slower. We fold reputation into the Reputable sub-score of the readiness stamp, Sybil-adjusted by client concentration so that a single wallet posting 500 five-star reviews for its own agent does not dominate. This is the emerging quality signal we expect to matter most over the next few reports — it is the first on-chain data point that lets you tell an agent that is used apart from an agent that is merely reachable. If you want to see how it factors in, check the AI Verdict block on any paid /v1/intel/agent response.

What this tells us five weeks in.

Three things stand out against Report 01. First, the raw registration count keeps climbing, driven overwhelmingly by BNB, and the total is now large enough that "how many agents are there" is no longer the interesting question. Second, endpoint churn is real and fast — a static top-N list is stale within weeks, which means the value is in an index that re-probes, not a directory that was accurate once. Third, the readiness distribution gives us, for the first time, a defensible answer to "which of these agents actually work" — and it is a small, coherent set. That is the shape of every other software ecosystem at this age, and it is the shape ERC-8004 is settling into now.

Every headline figure above is live from our public index — you can read the raw aggregates yourself at GET /v1/public/stats (free, no key), review the always-current ecosystem snapshot on the Ecosystem Stats page, browse the current rankings on the leaderboards, or pull a full enriched profile — owner identity, live tool inventories, Sybil-adjusted reputation, payment history — for any agent through the paid API.