The x402 tag holds 124 providers and 87 APIs. Its members average a Kin Score of 48.8 against 21.6 for the rest of the catalog, and an Agent Readiness of 39.2 against 12.2. Not one of the 124 is human-only. That is the headline, and it is real. The more useful number sits underneath it: 4.8% of these providers offer a dry-run mode, in a category whose entire premise is an agent paying for a call.
The five the tag ranks first are all exemplars. Agoragentic leads at 79.9, an agent marketplace and runtime settled in USDC on Base, carried by developer ergonomics at 85.7, and it is agent-native at 57.0 with an agent card, agent skills, idempotency, a dry-run mode and documented reversibility all lit. Hive Civilization scores 77.9 across 153 API pages of receipts, attestation and signature services, with a perfect 100.0 on access clarity, and lands agent-ready at 47.7. MandateShield also scores 77.9, a payment-authorization and fraud layer for agent transactions running three MCP servers, agent-native at 40.9. Delx scores 75.7 on 86.7 for discoverability, and at 55.4 lights fifteen agent dimensions yet stays agent-ready, not agent-native. Hilt also scores 75.7, a Solana and stablecoin checkout and subscription API, and posts the highest Agent Readiness in the five at 64.5, the only one with delegated identity and dynamic client registration lit, and the only one without an agent card. Every one of the five scores 31.8 on contract governance, the weakest facet each of them has.
Against the market, x402 is a different population. The median member scores 49.7 against a median of 14.8 for the rest of the catalog. 12.9% of the tag is exemplar on the Kin Score, against 1.9% elsewhere, and 22.6% is strong against 5.7%. No x402 provider is minimal, a band that holds 38.2% of everyone else. 83.9% run an MCP server against 14.6% of the rest, and 62.9% carry an agentic-access profile against 21.2%. The tag is not one thing, though: 66.9% of its members are crypto and DeFi payment providers, 11.3% are agent-native pay-per-call services, and 10.5% are agentic commerce, so the averages describe a crypto-heavy population more than a pure protocol cohort.
On agent readiness the shape is lopsided in an instructive way. 18.5% of the tag is agent-native against 1.8% of the rest, and 65.3% is agent-ready against 14.6%. The gaps are largest on the surfaces that tell an agent how to find and call a service: agent skills at 86.3% against 12.7%, error semantics at 95.2% against 22.7%, agent cards at 71.8% against 2.4%, auth clarity at 96.8% against 38.3%, rate-limit signals at 89.5% against 31.9%. Idempotency is lit for 37.9% of the tag against 4.8% elsewhere, a real lead. Then the gaps close. Dry-run mode is 4.8% against 1.4%. Reversibility is documented by 20.2% against 5.9%. Consent identity is 6.5%. Delegated identity is the one dimension where x402 sits below the market, at 10.5% against 13.1%.
The observation: x402 has built the door for agents and not the guard rail. This category has done the hard, visible work of being found, described and called by software, and it leads the catalog on almost every dimension that measures that. The dimensions it leaves dark are the ones that matter once the caller is spending money on its own: rehearse the call before paying for it, undo it afterwards, and prove which human it is acting for. Only 4.8% of x402 providers let an agent try a paid call without paying, and 10.5% can say whose authority the agent is spending under. The payments tag, measured here in September, was the one category that documented how to undo a call. The category that most wants agents to pay has not yet caught up to that.