Three things happened in the x402 ecosystem this week that no one is connecting. Together they tell a story about how the agent economy is actually being built — not top-down by enterprises, but bottom-up by developers shipping specs, plugins, and reference implementations.
On July 29, whawk46 submitted PR #2979 to the x402 specification repo: a discovery extension that standardizes how facilitators and resource servers announce themselves to the network.
The mechanism is elegantly simple:
_x402.<domain> points to a well-known manifest/.well-known/x402 (RFC 8615) returns a JSON document describing the facilitator's capabilities, endpoints, supported payment kinds, and optional attestation claims/supported — live data is always authoritative over the cacheThe TXT record format is minimal:
v=x402-1; wk=https://facilitator.example.com/.well-known/x402; k=usdc,usdt; net=base,solana; scheme=httpsAnd the well-known manifest resolves to something like:
{
"kind": "facilitator",
"baseUrl": "https://facilitator.example.com",
"endpoints": { "payment": "/pay", "receipt": "/receipt", "supported": "/supported" },
"kinds": ["usdc", "usdt"],
"networks": ["base", "solana"],
"assets": [{ "address": "0x833589fCD6eDb6E08f4c...", "network": "base", "symbol": "USDC" }]
}This is not just a GitHub PR. It has an accompanying IETF Internet-Draft: draft-hawkins-x402-dns-discovery, with IANA registrations filed for both the x402 well-known URI suffix and the _x402 DNS node name.
The model follows MTA-STS — DNS pointer → HTTPS authority, safe without DNSSEC. Curated directories become regenerable caches instead of load-bearing infrastructure. Any facilitator can become discoverable with one JSON file and one TXT record. Zero gatekeepers.
On August 1, Vohlsyr submitted PR #17541 to the ElizaOS registry: four third-party x402 plugins — @vohlsyr/plugin-concordance, plugin-trading-strategy-data, plugin-trustfetch, and plugin-chainscope.
These aren't toy demos. They underwent a security review that found a real signing flaw: the plugin clients were decoding untrusted x402 challenges from remote servers and signing arbitrary USDC authorizations with no local amount check. A compromised endpoint could drain an agent's wallet.
The fix — published as 0.3.0 on npm — adds assertChallengeWithinPolicy() before any signing: exact match on scheme, network, asset contract address, and payTo wallet, plus a per-request maximum ($0.25–$1.00). 55/55 adversarial tests pass.
This is real. ElizaOS — one of the largest agent frameworks — now has production x402 payment plugins with security-hardened signing. Agents built on ElizaOS can discover, pay for, and consume x402 APIs natively.
On July 18, substreambc opened issue #84 on the Hivemapper repo: "MCP x402 Pattern for Monetizing your valuable Hivemapper Data."
Hivemapper has geospatial data — road coverage, dashcam imagery, mapping telemetry — that AI agents are starting to consume. The proposal: standard MCP interface for agent consumption, HTTP 402 for payment settlement, on-chain receipts for auditability, BYO wallet with no custody.
The pattern is being copied from Helium MCP, which already does this for Helium network data. Real companies with real data are asking: how do we let agents pay for this?
Individually, each of these is a routine open-source contribution. Together, they form a pattern:
This is bottom-up standardization. No one at Visa, Mastercard, or Binance decided "now we shall standardize agent discovery." A developer wrote a spec, built a resolver, filed an IETF draft, and opened a PR. Another developer shipped x402 plugins for the largest agent framework. A third proposed the pattern for geospatial data monetization.
The agent payment stack is not being built by committee. It's being built by the people who need it.
Claude Code auto mode goes default on August 14 — 3 days from now. When that happens, Claude Code agents will have permission to spend money autonomously. They will need to answer three questions for every API they consider calling:
The IETF draft answers question #3. The DNS discovery mechanism means agent frameworks can ship with a resolver that finds x402 facilitators automatically — no hardcoded registry, no single point of failure, no gatekeeper. Just DNS.
Question #1 is what minia2a's trial-first + verify-first model already solves. Question #2 is what Aidress's trust scoring addresses. These are complementary layers of the same stack.
The discovery layer now has three competing philosophies, each backed by real money:
| Philosophy | Example | Funding | Model |
|---|---|---|---|
| API-key-first | AIsa.one | $6.5M seed | Humans provision keys, agents consume |
| Exchange-owned | Binance B402 | CEX treasury | Single-chain walled garden, gas-sponsored |
| Open-standard | x402 + IETF discovery | Community | DNS → well-known → live cross-check, zero gatekeepers |
The open-standard approach has no funding and no marketing budget. What it has is developers shipping specs, reference implementations, and IETF drafts — because they need this to work for their own agents.
The agent payment ecosystem is being built by the developers who need it, not by the enterprises that will eventually profit from it. An IETF draft for DNS discovery, production x402 plugins in ElizaOS, and real companies exploring MCP monetization — three independent signals pointing in the same direction.
Discovery is standardizing. Payments are being embedded in frameworks. Data providers are asking how to charge agents. The pieces are falling into place. August 14 is 3 days away.
minia2a is an agent-to-agent API marketplace with 306 x402 pay-per-call endpoints. Trial-first access (15 free calls per endpoint), live health probes every 60 seconds, facilitator-agnostic payments. Discover APIs →