Frequently asked questions
Agent402 is the open-source, self-hostable applied layer of Agentic Finance for x402 and MPP (+ MCP server): an open cross-seller Index, Smart Order Router, on-chain leaderboard and MPP marketplace, built on pay-per-call web tools for AI agents - free via proof-of-work, or paid in USDC on Base, Solana, Polygon, Arbitrum, Monad, Celo, Avalanche, Sei, Optimism, Stellar & Algorand - plus USDG on Robinhood Chain over x402, or over MPP (Base/Celo USDC, native Tempo).
Do I need an account or API key?+
No. Nothing here has a signup. Payment - USDC, proof-of-work, or a prepaid card-credits key you already hold - is the only credential, charged per call.
Can I pay by card instead of crypto?+
Yes: buy a finished report at /reports ($2 to $5 by card, refunded if it fails) or subscribe to a monitor at /monitors ($5 a month, cancel anytime). Prepaid card credits are not on sale; a credits key bought earlier still pays any tool with Authorization: Bearer a402_…, debited only when a call succeeds. The card price includes payment processing; an agent paying per call over x402 or MPP pays the lower tool price for the same report.
What does it cost?+
Flat per-call prices from $0.001. Most tools are $0.001 to $0.02; premium AI and media tiers run higher and multi-tool skill packs are $0.003 to $0.101; finished report products are $0.60-$2.00 per call for an agent per report ($2 to $5 by card at /reports, where the price includes payment processing). Every price is published in /api/pricing and quoted exactly in every HTTP 402 response. Calling a tool needs no subscription. The metered model route (/v1/metered) quotes each request from its body before payment and settles what the call used, under that quote.
Can I use it without any money or a wallet?+
Yes. Most pure-CPU tools accept proof-of-work - a sha256 puzzle solved by your own CPU (about 65,536 sha256 hashes on average, a fraction of a second of CPU on a laptop) - and the hosted MCP connector runs that same set for free (rate-limited).
What is x402?+
An open HTTP payment standard built on the 402 Payment Required status code, for machine-to-machine pay-per-call payments in stablecoins, with settlement infrastructure from Coinbase. Plain-English explainer: /what-is-x402.
What is MPP, and does Agent402 support it?+
Yes - every paid endpoint is dual-stack, and now natively via Tempo too. MPP (Machine Payments Protocol, the IETF-track Payment HTTP authentication scheme) carries the pay-per-call handshake through the web’s standard auth headers: the 402 carries a WWW-Authenticate: Payment challenge, the client pays via Authorization: Payment, and settled responses return a signed Payment-Receipt. Same URL, same price either way - MPP’s evm method settles identically to x402 (same on-chain USDC settlement), while its tempo method settles natively via Tempo’s own relay, a genuinely separate mechanism. How the two compare: /what-is-x402; the full MPP explainer: /what-is-mpp.
Which blockchain and asset does it use?+
USDC on Base (or Solana, Polygon, Arbitrum, Monad, Celo, Avalanche, Sei, Optimism, Stellar, Algorand - or USDG on Robinhood Chain). The buyer needs only the stablecoin - gas is sponsored by the facilitator on EVM chains.
Does using this spend my AI tokens?+
No. Not your tokens, ever: you never hand us a model key, and a model we run is billed to us and priced into the call. Most of the catalog is deterministic code (parsers, hashes, math, a real browser) with no model in its path. The ones that do run a model are named: the /v1 gateway tiers, the report products, Decide (/api/decide), the judgment model the Smart Order Router uses to pick among shortlisted tools, and the image, speech, transcription, embedding and AI-answer tools. Proof-of-work spends your CPU; x402 spends USDC.
Is there an OpenAI-compatible endpoint?+
Yes - /v1 is a pay-per-call OpenAI-wire LLM gateway: point any OpenAI SDK at base_url https://agent402.tools/v1 for chat (flat quality tiers, a metered tier quoted per request, model-optional auto-routing), embeddings, and image generation. No API key, no signup - settle in USDC over x402, same as every other tool. See /pricing for the tier breakdown.
Is my data stored?+
Tool inputs are processed in memory to compute the response. The memory tools store data by design (wallet-keyed, owner-deletable, with optional TTL). An input is otherwise kept only in the cases the privacy policy lists, such as short-lived response caches, catalog searches that find no tool, decisions bought on /api/decide (30 days), and the feedback you leave on a paid call. Full policy: /privacy.
How do I know the service is honest?+
It is fully open source; CI re-tests every non-metered tool against its own documented example before each deploy (the metered ones - LLM tiers, report products, search and the tools that need a third-party key - are excluded so CI never spends upstream); and revenue settles on-chain to agent402.base.eth (the named public receiving wallet), auditable by anyone on Basescan.
What happens if a tool fails after I pay?+
You are not charged. Payment settles only for a successful (under-400) response, so an error cancels settlement and no money moves. On top of that guarantee, anything which can't be served reliably is removed from the catalog rather than left to fail, and failure rates are watched by CI and a 15-minute production heartbeat.
Is Agent402 self-hostable and open source?+
Yes - the server is open source under the AGPL-3.0 license (the client SDK, MCP connector, and tollbooth are MIT). Clone the repo and run it yourself for free, with or without payments enabled. It also ships agent402-tollbooth, an open-source pay-per-crawl gate for charging AI crawlers on your own site.
Can I find tools on other x402 sellers from here?+
Yes. Agent402 is also an x402 Index + Smart Order Router: POST /api/route ranks tools across every x402 seller we have crawled - the local catalog plus sellers auto-discovered from public registries like the Coinbase CDP Bazaar, which are re-read every hour, with every known seller re-probed every 30 minutes. Sellers whose recent crawls errored are filtered out. Candidates are shortlisted by how well they match the task, then ordered by crawl health, by distinct payers over the last 30 days, by price, by a discovery listing's curated flag and by slug length, shorter first; a judgment model can then pick the one that does the job from that shortlist, or decline them all, and when it is not consulted the shortlist order stands. Browse the live marketplace at /marketplace or fetch the index as JSON at /api/index, which is paginated - one page at a time, with ?seller=<host> for a single origin. Both surfaces are free.
How do I list my own API?+
For free, three ways: your origin is auto-discovered from public x402 registries (Coinbase CDP Bazaar, GoPlausible) once it’s live and settling; paste it on /sell for an immediate probe; or call POST /api/index/register directly. A listed seller is routable by the Smart Order Router and ranked on /leaderboard by real on-chain USDC volume - 0% take, settlement lands straight in your wallet.
How do I see which x402 sellers are most used?+
GET /api/leaderboard returns the head of the live on-chain ranking of x402 sellers by Base USDC settled volume - callsSettled, totalUsd, and uniqueBuyers per seller. It is a top-N slice, not the whole board: 25 rows by default, up to 50 with ?top=N, and the response carries totalSellers for how many are ranked in all, so a seller absent from the rows you were served may still be ranked below them. The pipeline walks every page of the Coinbase CDP Bazaar discovery endpoint, queries eth_getLogs on Base USDC for each seller’s payTo, counts a transfer that matches a price the seller publishes and otherwise holds it to the per-call ceiling the response reports as maxCallUsd (larger inbound is funding/swaps, not buys), and aggregates over the window the response reports as windowServed. The snapshot refreshes hourly server-side. Free, like /api/find and /api/route. Use ?include=external to exclude Agent402 itself and rank only the rest of the ecosystem.
How does the Smart Order Router decide which seller to route to?+
Candidates are shortlisted by how well they match the task, then ordered by crawl health, by distinct payers over the last 30 days, by price, by a discovery listing's curated flag and by slug length, shorter first; a judgment model can then pick the one that does the job from that shortlist, or decline them all, and when it is not consulted the shortlist order stands. Seller health is computed from the last five crawl outcomes. If none of them does, route-and-execute refuses before anything is paid. Sellers whose recent crawls errored are excluded entirely - a buyer routed to a dead seller wastes money. Brand-new sellers with no history yet are still routable: benefit of the doubt for newcomers.
Who runs Agent402?+
Havok Holdings LLC - a public, contactable maintainer reachable at mike@agent402.tools, on GitHub, and on X.