LetAgentPay

LetAgentPay is an open-source system for controlling how AI agents spend money. It gives an agent a wallet for paying for APIs, services, and tasks, while applying budgets and policies set by the user. The site describes it as working with both fiat and crypto rails, and as supporting USDC on Base via x402.

Its core design centers on one policy engine that sits between an agent and a payment rail. Before a purchase goes through, the agent requests approval with the amount, category, and description, and the engine runs checks on budget, category, schedule, and transaction history. The product supports budgets at multiple levels, including per-request, daily, weekly, monthly, and overall limits; category controls with custom per-account categories; and schedule-based rules such as business hours or weekend blocking. It can auto-approve routine purchases, escalate unusual or high-value requests for manual review, and keep a full audit trail of every action. The page also says rules can be written in plain English or JSON, with Claude converting natural-language policies into enforceable JSON.

LetAgentPay includes a real-time dashboard for seeing requests as they happen and approving, rejecting, or reviewing them with full context and policy results. Integration options listed on the site include Python and TypeScript SDKs, an MCP server, a REST API with bearer token authentication, and guides for nine frameworks. The MCP server is said to work with Claude Desktop, Cursor, and other MCP-compatible AI tools. The product is described as self-hosted and open-source, with full source access, and also available as a hosted SaaS that uses the same engine and APIs.

The page says it is built for agent developers, teams running AI in production, and builders working with x402 or AP2. It is free during early access, and the site includes prompts to create a free account and view the documentation.