agentic-pay is a cross-platform, non-custodial payment engine for AI agents — MCP-native, with built-in spend limits and human confirmation, across seven crypto rails.
Wallets, key storage, and spend controls get bolted on ad hoc, differently, every time — and most of those implementations have no real limit on what an agent can spend.
Every agent framework that wants to transact builds its own wallet integration from scratch — inconsistent, unaudited, and rarely covering more than one or two assets.
A hallucinated tool-call or a compromised agent can move funds with no ceiling. Spend limits and human-confirmation thresholds are usually an afterthought, if they exist at all.
Every send — from MCP or REST — passes through the same policy and idempotency layer before it ever touches a chain.
This is the actual product: not the ability to send a transaction, but the guardrails around letting an autonomous agent do it unsupervised.
Hard, non-negotiable ceilings per agent, enforced before any send reaches a rail adapter.
Payments above a configurable threshold are rejected until a human explicitly reviews and confirms — never settable by the agent itself.
Every send requires a unique key. A retried tool-call can never double-pay.
Your keys, your infrastructure. agentic-pay is a payment engine, not a custodian.
Restrict any agent to a fixed set of approved destinations, optionally, per policy.
Every invoice and payment is recorded locally, queryable by agent, by asset, by time window.
Pro and Managed tiers are in development — get in touch below for early access and to help shape them.
For Pro/Managed early access, Enterprise licensing, or commercial licensing inquiries — reach out directly.