About
How x402route works.
A discovery service for x402-paid endpoints. Agents send a task and budget; we return one recommendation backed by live uptime data and 30-day usage signals. Below: the principle of how we pick, what feeds the catalog, and what we deliberately do not do.
Selection principle
Filters before scoring.
filter 1
Chain match
Endpoints that don't support the requested chain are removed first. The chain is part of the request, not a hint.
filter 2
Budget
Anything above max_price_usd is dropped before scoring runs. We never recommend over-budget out of "quality".
filter 3
Hard liveness
Endpoints with zero successful pings over the last 24 h are excluded entirely. A confirmed-dead endpoint can never win.
scoring
Three signals
Among survivors we weigh popularity (real on-chain x402 settlements in the last 30 d, counted directly from Base mainnet), payer diversity (unique buyer wallets), and current uptime. Soft red-flag penalties apply for shallow descriptions and synthetic examples — unless the endpoint has earned popularity to override them.
mode
Agent intent
The mode (cheapest, balanced, best_quality) decides how the score and price combine for the final pick. Cheapest excludes zero-score endpoints to avoid unsafe budget picks.
always
Inspectable
Every per-task score is reachable through GET /v1/score?endpoint=URL. If you don't agree with a pick, you can audit it.
Where the catalog comes from
Two catalogs, three signal feeds.
Daily catalog pull
We sync both Coinbase Bazaar (~46K resources) and agentic.market (~700 services), dedupe by URL, and apply spam filters before adding to the active set.
Live pings
Every monitored endpoint is pinged on a 30-minute cycle. Status, latency, and alive/dead are stored as a rolling 24-hour window — that feeds the uptime filter.
On-chain quality
Daily we count actual USDC settlements (filtered by EIP-3009 selector) to each endpoint's pay_to address on Base mainnet. This is our popularity signal — independent of self-reported metadata.
Curated layer
On top of raw catalogs, a hand-picked map of task → endpoints drives recommendations. Spam, MCP-only and dead endpoints are excluded.
Probe-based pay_to extraction
Daily we POST each endpoint to capture the live 402 challenge. The returned payTo beats stale catalog metadata; if probe fails, we fall back to catalog-published value.
On-demand refresh
Asking /v1/score?endpoint=URL for an endpoint with quality data older than 24 h triggers an immediate single-endpoint on-chain refresh. Always fresh when queried.
Boundaries
What we don't do.
Not a data API
The response is a recommendation: which endpoint, why, alternatives. The actual token price, RPC result, or news payload is the downstream provider's job.
No proxying
We don't relay calls or hold credentials. Agents pay the recommended endpoint directly through x402. We don't see the payload.
No accounts
No login, no API key, no subscription. Each POST /v1/route is settled with a $0.001 USDC payment over x402 on Base mainnet. That's it.
FAQ
Quick answers.
What if the recommended endpoint goes down right after?
Our hard uptime filter removes endpoints with zero pings in 24 h. Drift between a recommendation and a downstream call is on the order of seconds; a recommendation older than that should be re-fetched. The response also includes alternatives for immediate fallback.
How do I get my endpoint into the catalog?
Publish it on Coinbase Bazaar or agentic.market (we sync both daily). Once we observe x402 settlements to your pay_to on Base mainnet, you'll show up in our scoring. Curation into a specific task happens after we see the endpoint stable.
Why $0.001?
Low enough that an agent calling 100 endpoints per workflow pays cents in routing overhead, but non-zero so spam is naturally throttled. Settlement happens on Base, where fees are negligible.
Are scores final?
No. Popularity, diversity, and uptime change continuously. Two requests for the same task one hour apart can return different winners — that's the point.