NOTE: re-routing reply → article because source=other (Stripe blog, no reply thread, email=null). product_fit=mnemopay, score=94 ≥ 85. Framing as a developer analysis piece, not a competitor attack.

Stripe's Machine Payments Protocol — what it covers and what it doesn't

Stripe shipped MPP at Sessions 2026. Session-based streaming payments, no accounts, no subscriptions, Stripe's compliance stack built in. Agents can shop, pay for reservations, buy tickets — on the rails that already handle most of global e-commerce.

That's a meaningful launch. Stripe's compliance infrastructure is real, their fraud detection is real, their global network is real. MPP is the right primitive for agents doing consumer-style purchasing within Stripe's ecosystem.

The two things MPP doesn't address are the two things that matter most as agents get more autonomous: dynamic trust scoring across the agent's full payment history, and protocol-agnostic integration for teams that can't or won't be Stripe-exclusive.

what MPP actually ships

Session-based architecture: each agent session gets a payment context with configured limits. The agent can transact autonomously within those limits without additional approvals. Stripe's compliance stack (fraud detection, sanctions screening, dispute handling) applies automatically.

The "no accounts, no subscriptions" model means a developer can spin up an agent session, wire in MPP, and have a payment-capable agent in an afternoon. That's genuinely low-friction. The developer experience here is a strength.

the trust scoring gap

MPP's session limits are static: you configure the limit at session creation. A new agent session and a session with 10,000 prior successful transactions get the same limit ceiling for that session type.

That's correct behavior for Stripe's model — they're a payment network, not a creditworthiness scoring system. But for developers running agents at scale, the missing piece is a dynamic trust signal: as an agent accumulates a clean transaction history, it earns a higher ceiling. As it shows anomalous behavior, its ceiling drops automatically.

Agent FICO (300–850) fills this gap. The score is computed from transaction history across sessions — not per-session, but across the agent's full operational history. An agent that's processed 50,000 transactions cleanly over six months has a different risk profile than a new agent. Static session limits can't capture that.

The Agent FICO integration with MPP: the policy evaluation step runs before the Stripe session is created. If the agent's FICO score qualifies for a higher session limit, the session is initialized with that ceiling. If the agent is new or has anomaly signals, the session starts at a conservative limit. The FICO score is the input; the MPP session limit is the output.

the protocol-agnostic case

MPP is Stripe's protocol. x402 is Coinbase's protocol. A2pay is the A2A Protocol's payments component. Seven protocols have shipped since February 2026.

Developers building for multi-rail environments can't bet exclusively on one protocol. Enterprise compliance requirements may require fiat, crypto, or both. International deployments may require payment rails Stripe doesn't cover. Future standards may supersede any of today's protocols.

MnemoPay's SDK is transport-agnostic: the Agent FICO scoring, policy evaluation, and signed receipt chain run independently of the payment rail underneath. You wire in MPP for Stripe transactions, x402 for crypto, whatever comes next when it ships — the trust layer stays stable.

672 tests cover the multi-rail abstraction, ensuring the Agent FICO evaluation produces consistent results regardless of which rail the payment resolves on.

the developer stack for session 2026

The practical stack most enterprise agent developers are building toward:

• Stripe MPP for consumer-style payments on fiat rails (Stripe's network, their compliance)
• x402 / AgentCore for crypto rails and AWS-native deployments
• Agent FICO + MnemoPay for the trust layer that spans both — dynamic limits, anomaly detection, signed receipts, protocol-agnostic

MPP is table stakes for agents doing e-commerce. The trust layer is what you add when you're running agents with real budget authority across multiple payment protocols at scale.

MnemoPay v1.0.0-beta.1, 1.4K weekly npm downloads: https://mnemopay.com