x402 & Agent Payments
x402 is an open payment protocol built into the request itself. For us, it is the likely rail for agents buying EIPs that are still ahead of the open hardfork.
It is not part of the open server. The hosted MCP is free. Payment follows after we have usage, a stable service, and something new to sell.
The protocol can pay today. The agent experience is still being assembled.
What x402 is now
HTTP already had a status for payment: 402 Payment Required. x402 defines the request, payment proof, verification, settlement, and retry around it. A facilitator can verify and settle onchain without the seller running payment infrastructure.
100M+
Payments reported by x402 since May 2025
V2
Modular protocol, MCP support, discovery extensions
50+
Organizations in the Linux Foundation x402 Foundation
That makes it a natural fit, not an automatic choice. We still need to prove the demand, the price, and the wallet path.
Primary references: x402 V2, x402 Foundation, protocol documentation.
Why it fits this product
| Property | Fit | Question still open |
|---|---|---|
| Pay per call | A fixed USDC amount per tool, known before the run | Whether two and four cents hold once people use it |
| No account required | Agents can buy one capability without a SaaS signup | What receipt and support path replaces an account history? |
| Stablecoins | Our early users already understand wallets and Base | How much wallet setup will they tolerate? |
| Machine-readable quote | The buyer can decide before EthereumJS runs | Which facts must every quote expose? |
| Open ecosystem | Wallets, facilitators, and sellers can evolve separately | Which providers are reliable enough for production? |
The product remains the deterministic run. x402 is the access rail. Pricing owns the free-to-paid cycle; token utility owns any later holder discount.
The integration gap
x402 V2 already has MCP wrappers, buyer spend controls, lifecycle hooks, discovery through Bazaar, and signed offers and receipts. Coinbase ships a separate Agentic Wallet MCP with funding and human-set limits. Open-source wallet MCPs can also be added to Cursor and Claude.
What does not exist yet is one native wallet contract shared by Cursor, Claude, and every other harness.
| Layer | State | Consequence for us |
|---|---|---|
| Seller protocol | Usable now | We can build standards-first on x402 V2 |
| Settlement | Hosted facilitators exist | Keep the facilitator replaceable |
| Wallet | Working companion MCPs exist | Document paths; do not build custody |
| Harness | Wallets are add-ons, not a native universal slot | Expect setup, funding, and policy friction |
| MCP authorization | OAuth protects access; it does not provide payment | Keep authentication and payment separate |
| Autonomous policy | Spend caps and hooks exist in code | Never rely on a model prompt to protect funds |
Integration references: x402 with MCP, buyer spend controls, Coinbase Agentic Wallet MCP, MCP authorization.
Use the early audience
Our likely design partners are unusually technical and crypto-native: protocol engineers, auditors, wallet teams, agent builders. The setup cost can become part of the offer.
The message is not “pay because x402 is exciting.” It is: get useful early protocol access, and help make agent-native payment real.
What we can offer:
- a small paid beta around frame transactions and the EIPs after it;
- an explicit compatibility matrix for wallets, harnesses, and facilitators;
- short, public reports on setup time, failed payments, retries, and receipts;
- direct feedback into quote semantics and buyer policy.
Agent-to-agent horizon
The goal is a buyer agent that can discover the service, judge the offer, pay within policy, run the tool, and verify what came back—without a person reading this site.
Prepare now
Seller contract
- Stable server identity, tool names, schemas, and versioning
- Quote without execution; no paid work before authorization
- Idempotent retry and replay protection
- Facilitator adapter rather than provider logic in the engine
- Receipt linked to the tool call and its provenance
- Bazaar-ready discovery metadata for the paid tools
Decide before the paid beta
| Decision | Plausible path |
|---|---|
| Charge shape | Fixed price per tool (Pricing). Exact quotes, not metered gas |
| Buyer policy | Amount, asset, network, seller, tool, per-call and session ceilings |
| Failure | Define expired quote, rejected signature, settlement timeout, engine error, and retry |
| Delivery proof | Signed offers and receipts, plus the existing run provenance |
| Discounts | Apply outside the engine; never make the community token a prerequisite |
| Discovery | Bazaar first for x402; MCP registry later for general discovery |
Three to five months
Likely: more wallet MCPs, more Bazaar-listed tools, stronger spend policy, and better receipts. Possible: harnesses expose wallet UX more directly. Uncertain: one wallet interface becomes standard across them.
Our position should work in all three cases:
- Seller-ready. A conforming x402 buyer can pay us.
- Client-agnostic. No dependency on one wallet or harness.
- Safe by construction. Buyer policy lives in code; retries cannot double-charge.
- Discoverable. Metadata is ready when registry work moves from Later.
- Observable. We can see quote → payment → execution → receipt without storing agent secrets.
The first milestone is not “fully autonomous commerce.” It is one external agent, using a bounded wallet, discovering and buying one useful run end to end.
Changelog
- v0.82026-10-06x402 stays off the open server. The hosted MCP is free.
- v0.72026-10-05Charge shape aligned with Pricing: a fixed amount per tool, not metered gas.
- v0.62026-10-05Rebuilt around x402 maturity, the harness-wallet gap, crypto-native design partners, and the decisions needed for agent-to-agent payment.
- v0.52026-10-01x402 moves to the post-launch paid tier (EIP-8141 first). Launch week is the open Glamsterdam MCP.
- v0.42026-09-24402 quote as product surface; discovery via registries and other agents.
- v0.32026-09-02x402 decided for launch week (USDC on Base) — rails not hook; facilitator wiring build-in-public.
- v0.22026-09-02Oracle built — x402 framed as launch-week payment rails, not the product hook; MCP tools exist.
- v0.12026-06-30Initial outline — x402 payment flow, MCP as delivery layer, registry discovery.