Payment infrastructure for AI agents.
MPPCash gives every AI agent a wallet, a spending policy, and an on-chain transaction record — settled in USDC on Solana, governed by the Machine Payments Protocol (MPP).
Think Stripe for machines: one line to integrate, instant to settle, fully auditable.
| Repository | Description |
|---|---|
app |
TypeScript and Python SDKs — wrap any HTTP client for automatic MPP payment handling |
smart-contracts |
Open-source Solana programs: Policy Program and Settlement Program |
api |
REST API for wallet provisioning, balance, policy management, and transaction history |
documentation |
Documentation site — quickstart, API reference, guides, architecture |
AI agents call paid APIs every day. When a service needs payment, it responds with HTTP 402. MPPCash handles the rest.
Agent makes request
→ Service returns HTTP 402 + payment challenge
→ MPPCash SDK checks wallet balance + active policy
→ If policy allows: signs payment credential, retries request
→ Service fulfills request, returns receipt
→ MPPCash batches receipts into periodic Solana settlement
The agent's application logic sees none of this. It makes an HTTP request and gets a response.
Agent Wallets — Every agent gets its own wallet, provisioned via API in under 10 seconds. Wallets hold USDC on Solana and are identified by a Solana public key and a human-readable handle (@my-agent.mppcash). Custodial by default; non-custodial if you bring your own keypair.
Spending Policies — Every wallet is governed by a programmable policy enforced on-chain by the Policy Program. Policies are not UI guardrails — they are Solana program accounts that revert transactions that violate them. A policy can set per-transaction caps, time-bounded budgets, allowlists, category restrictions, and human escalation thresholds.
On-Chain Ledger — Every settlement is a Solana transaction. It is queryable via the MPPCash API and exists independently of MPPCash. The ledger cannot be edited. Compliance teams can audit exactly what each agent spent, when, on what, and against which policy — with cryptographic receipts signed by both parties.
TypeScript
import { MPPCash } from '@mppcash/sdk';
const agent = new MPPCash.Agent({
wallet: process.env.MPPCASH_WALLET_KEY,
policy: 'default',
});
const client = agent.createHttpClient();
const result = await client.get('https://api.someservice.xyz/inference');
// Automatic HTTP 402 handling — no other changes neededPython
from mppcash import Agent
agent = Agent(
wallet=os.environ["MPPCASH_WALLET_KEY"],
policy="default",
)
response = agent.get("https://api.someservice.xyz/inference")
# Automatic MPP payment on 402 responsesFull documentation at mppcash.xyz/docs.
MPPCash implements the Machine Payments Protocol (MPP) — the open standard for agent-to-agent and agent-to-service payments, with 100+ compatible service providers including Anthropic, OpenAI, Alchemy, Shopify, and Dune.
Settlement runs on Solana: sub-penny fees, ~400ms block time, 65k+ TPS. USDC is the primary settlement token. There is no MPPCash token.
For high-frequency agents, MPP Sessions let agents pre-authorize a spending cap once and stream micropayments off-chain continuously. Thousands of micropayments aggregate into a single on-chain settlement — Solana fees stay negligible at any throughput.
| Framework | Status |
|---|---|
| LangChain / LangGraph | Beta |
| OpenAI Agents SDK | Beta |
| CrewAI | Beta |
| Vercel AI SDK | Beta |
| Component | Trust required in MPPCash? |
|---|---|
| Spending policy enforcement | No — enforced on-chain |
| Settlement correctness | No — verifiable on Solana |
| Wallet custody (custodial mode) | Yes |
| Off-chain micropayment tracking | Yes — during session, before batch settle |
| SDK correctness | Yes — open source, audited |
Non-custodial mode is a first-class option, not a roadmap item.
MPPCash is in private beta. The on-chain programs and SDK are open source. APIs and program addresses are subject to change during beta.
Built on MPP. Settled on Solana. No speculative token. No dark patterns.